{
  "schema": "signal-return-gallery-v1",
  "builtAtUnix": 1791563859.7345395,
  "selection": {
    "seed": "577e3090670be99d3c929a7aa1e7b555",
    "selected_at": 1791562693.7951193,
    "method": "Five uniformly sampled distinct clip ranks per corpus among jobs committed in all three codec shard sets. No listening or content-based screening. Original means the exact canonical tokenizer input; upstream-enhanced inputs are labeled.",
    "populations": {
      "iptv": {
        "eligible_jobs": 464419,
        "eligible_clips": 2738176,
        "drawn_ranks": [
          1551413,
          644600,
          331566,
          1395086,
          185962
        ]
      },
      "podcast": {
        "eligible_jobs": 102876,
        "eligible_clips": 5401616,
        "drawn_ranks": [
          3805698,
          1076122,
          4115633,
          1515487,
          1468430
        ]
      },
      "h3": {
        "eligible_jobs": 4704,
        "eligible_clips": 1569650,
        "drawn_ranks": [
          290375,
          56484,
          228728,
          156840,
          52973
        ]
      }
    }
  },
  "selectionManifestSha256": "cd040bda784d0d34333499f79a110dda59aea57b65d3b349553902c60fef6ca3",
  "decoderInputManifestSha256": "710317bf53ccc20f3359429aea41048d053a4dd9d2c85823e254f57cefbeef65",
  "models": {
    "qwen": {
      "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
      "model_revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
      "checkpoint_files_sha256": {
        "model.safetensors": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
        "config.json": "ee65bb901c876664ab8707c487157aa1a6ee57c65969b28fb5ec9dc211e68167",
        "preprocessor_config.json": "fcb3805e597e786d4067706e602f6688524640f8d3396790e2e09b5942fcbdfb",
        "configuration.json": "6bc26d64eb5024b4d1dab5a52371958b429256d6c9d59787f1f5294a54e0cebd"
      },
      "official_source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
      "decoder_tool_sha256": "86cf56c58960da84be03d98a22d2cfb11d6c6d88ff793b5063076642c9d7f203",
      "precision": "float32; matmul/cuDNN TF32 disabled",
      "decoder_api": "Qwen3TTSTokenizer.decode([{'audio_codes': saved_codes_T16}])",
      "native_decoder_clamp": [
        -1,
        1
      ]
    },
    "chatterbox": {
      "model_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
      "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
      "tokenizer_checkpoint_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
      "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
      "decoder_script_sha256": "1b478dfceca2a9b4cb43f189257b96145393073e153425e44d5584d1eb7a9791"
    },
    "fish": {
      "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
      "decoder_sha256": "a77bd97082588202dc100c817cb0661bf8ede58e7132702663becf235f07a69c",
      "source_sha256": {
        "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
        "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
        "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
        "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
        "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
        "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
        "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
        "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
        "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
        "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
        "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
      },
      "precision": "reference;float32;cuDNN TF32 enabled;matmul TF32 disabled;no autocast",
      "method": "bundled quantizer.from_codes(all10books) -> bundled model.decode",
      "normalization_scope": "Production Fish tokens encode a per-clip -16LUFS-normalized waveform. Original loudness is not encoded in these ten codebooks; decoder output retains model gain."
    }
  },
  "decoderReportSha256": {
    "qwen": "c98c0a7e9a5cb78ffacfd342193dd7d652171fa873fc608aa011573058c79c11",
    "chatterbox": "8479c602108fb987189ef36106a9adfb558818f0e3101650d29724701083a5ee",
    "fish": "bee6e19ee90d5cdd0ffb38b72c12ae457781fb61e44a5ddfbae1629d2d1e9a52"
  },
  "policies": {
    "contentSource": "Previously encoded and committed R2 token shards; no fresh encoding.",
    "original": "Exact canonical mono float32 encoder input, verified against saved PCM SHA256. Upstream-enhanced source variants are labeled.",
    "chatterbox": "Conditional reconstruction: source speaker/mel reference up to 10 seconds, prompt prefix from saved tokens. No reference tokenization. Official decoder fade retained.",
    "fish": "Production Fish tokens encode a per-clip -16LUFS-normalized waveform. Original loudness is not encoded in these ten codebooks; decoder output retains model gain.",
    "output": "Unmodified raw native-rate FLOAT WAV, all frames retained. No gain matching, extra clipping, or alignment shifts. Native frame rounding can make S3 output slightly shorter than its original; exact duration deltas are recorded.",
    "playback": "Shared original timeline, playback stops at original duration; downloads preserve frame padding. Browser resamples to device rate. One global volume control.",
    "waveforms": "128 peak bins on the original time window; one common amplitude scale per clip.",
    "scope": "Random listening sanity check, not a quantitative evaluation or representativeness claim."
  },
  "clips": [
    {
      "id": "iptv-01",
      "corpus": "iptv",
      "run": "english-35kh-20261006",
      "jobId": "90fd37ce-70e3-585d-aba4-5bd6f1c8fa22",
      "sampleId": "fadf5d51c88d8dc721836060df30200dcf4011c4fe1299b2e0bdd4d34b2222be",
      "selectionRank": 1551413,
      "inputVariant": "original",
      "source": {
        "uri": "r2://iptvdata/CNBCe.tr/1786824815.aac.s555enc",
        "objectSha256": "3220289ff26afec1ff880400df597615c3df9f25ddb3beb0ac09369a11519952",
        "startSample": 11072352,
        "sampleRate": 48000,
        "numSamples": 221184,
        "pcmSha256": "d2a5939d8f38a4736c61c31dae62adc79a65cf3ea9a0d1475d023f0f9f981ef2",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/iptv/shard-1f82e92f6d0202b1117ca247b9c656af.npz",
          "shard_sha256": "701b574f433da0ec4989bd9e0584185eefb266228ff845e01236fbf8f8552026",
          "record_index": 58026,
          "sample_id": "fadf5d51c88d8dc721836060df30200dcf4011c4fe1299b2e0bdd4d34b2222be",
          "token_shape": [
            16,
            58
          ],
          "tokens_npy_sha256": "4f9b7ffd4c853763d4a278b60cc2d97128f41261ed177058dc28d2a6e30bbea5",
          "tokens_payload_sha256": "3e6a8bc3d810974ac4314e8fea2e676200f848f93ffe8ffe774fa4c298ce5949",
          "downloadedAtUnix": 1791562833.661574,
          "shardBytes": 497297451,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 768,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/iptv/shard-b3a2f2ef329355fc524367390ea2c81f.npz",
          "shard_sha256": "ca370349bcc219cbb9ac40df01f815cacfae33fca31e3508edc1a0291f070e96",
          "record_index": 204855,
          "sample_id": "fadf5d51c88d8dc721836060df30200dcf4011c4fe1299b2e0bdd4d34b2222be",
          "token_shape": [
            1,
            115
          ],
          "tokens_npy_sha256": "c49eba86bb896fc1f9b5c7986edd816ea5b4c35cffee4ab61ee7d5a92e227b88",
          "tokens_payload_sha256": "c672d8f1e250b974b3cfb290d7337f539b32559a22651848bce86bbc7d2223d8",
          "downloadedAtUnix": 1791562829.9378345,
          "shardBytes": 497632411,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 4.6,
              "num_samples_24k": 110400,
              "stored_prompt_tokens": 115,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/iptv/shard-1c3ac292a605361cb3952c821e70b867.npz",
          "shard_sha256": "a466561dfbe2fe12ba8f0e78c5d347758d91fcd580e56b8fc2a25d19568987a9",
          "record_index": 57226,
          "sample_id": "fadf5d51c88d8dc721836060df30200dcf4011c4fe1299b2e0bdd4d34b2222be",
          "token_shape": [
            10,
            100
          ],
          "tokens_npy_sha256": "c4d22471045808555fdf0e4d0a2f7b82601c5ed7053e3927e8b5fc9a5f1259fb",
          "tokens_payload_sha256": "2fbb1fdf02f6e6c8fc61f6fba616f6b1de571fc2e63633a503010d02f72ee542",
          "downloadedAtUnix": 1791562836.375267,
          "shardBytes": 498117399,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 203212,
            "reference_verification": {
              "acoustic_books": 9,
              "codebooks_used": 10,
              "decoder_only_warmup": {
                "calls_discarded": 1,
                "cold_versus_warm_exact": false,
                "cold_versus_warm_max_absolute": 0.0011268854141235352,
                "cold_versus_warm_rms": 7.263506493649083e-05,
                "encoding_performed": false
              },
              "decoder_stage_output_exact": true,
              "post_module_upsampled_latents_exact": true,
              "projected_latents_exact": true,
              "semantic_books": 1
            }
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 111360,
          "duration": 4.64,
          "sha256": "c0fd7cbe877351cc3720c08161edb27d555a4335beb5c0908f62cac9c1b55c0f",
          "pcmSha256": "8194d6bea0b527e116d6d536d20ae84d0a8acd8f96f7749c40aae3eda5ad70b9",
          "peak": 0.47728586196899414,
          "rms": 0.06401089293481292,
          "durationDeltaSeconds": 0.03200000000000003,
          "url": "audio/iptv-01.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 110400,
          "duration": 4.6,
          "sha256": "602ff08127ed99a68d3dddcb203807622b91ed1a57ac615fa115cc173e74dcb2",
          "pcmSha256": "c1a9d2f0b9c32a83ff45b5760d5cbedda4a24ce9b228384eeda85b33f7bbcf2d",
          "peak": 0.5582633018493652,
          "rms": 0.07106229345850817,
          "durationDeltaSeconds": -0.008000000000000007,
          "url": "audio/iptv-01.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 204800,
          "duration": 4.6439909297052155,
          "sha256": "16a84952ae4c6f679ae03f086e6c82f745fdc1dcb45bcb37c4f5a2dca4809708",
          "pcmSha256": "40ecfce6c4edd26f54447ff38f26bcc56582a60de86f698a4b76efe55ff95bb9",
          "peak": 0.9991758465766907,
          "rms": 0.14673369240593453,
          "durationDeltaSeconds": 0.03599092970521589,
          "url": "audio/iptv-01.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 221184,
          "duration": 4.608,
          "sha256": "273aae1793183d3ef35c60b8a1b2e38b0192cb5533c88a81d50d3aa40e97c079",
          "pcmSha256": "d2a5939d8f38a4736c61c31dae62adc79a65cf3ea9a0d1475d023f0f9f981ef2",
          "peak": 0.4731728434562683,
          "rms": 0.06772703910564082,
          "url": "audio/iptv-01.original.wav"
        }
      }
    },
    {
      "id": "iptv-02",
      "corpus": "iptv",
      "run": "english-35kh-20261006",
      "jobId": "3c673e71-db84-5496-ae6b-2bb281e8db3f",
      "sampleId": "1f1a0cf61233ed36a5867777f0de217fa3b0481dfa6bfd86eb0b4e5bde7b7cc5",
      "selectionRank": 644600,
      "inputVariant": "original",
      "source": {
        "uri": "r2://iptvdata/CruzTV.us/1787575393.aac.s555enc",
        "objectSha256": "d31e322e5bb21ba3bd56a1ada77351111315b25738d0d6518dba504f1171b1a7",
        "startSample": 8395488,
        "sampleRate": 48000,
        "numSamples": 172032,
        "pcmSha256": "eebc0fe5435aadc050b34d6b245d02b84d7bb21015f796fb2c51500485969f39",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/iptv/shard-faac41bbb4e2827c96f7fe69aa0d8262.npz",
          "shard_sha256": "2bedd5cef713f468d2ae6ba915c072afb00ca378db7566473b2617b4f817020a",
          "record_index": 26697,
          "sample_id": "1f1a0cf61233ed36a5867777f0de217fa3b0481dfa6bfd86eb0b4e5bde7b7cc5",
          "token_shape": [
            16,
            45
          ],
          "tokens_npy_sha256": "111f041524f615d1ccf3daa0d34e564ba2de6531c0e81c3791ee9010487814d7",
          "tokens_payload_sha256": "2bb1d8e22bca71b9f5975b3f9eece470565ad8a4622881352641084bbe8781ce",
          "downloadedAtUnix": 1791562811.176481,
          "shardBytes": 497126130,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 384,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/iptv/shard-df020a13848644f3ce03886b5feaac20.npz",
          "shard_sha256": "3c0a9b159532714213222d7f95e2fbeee8dc714cbd83f077f06e16a323b8c05b",
          "record_index": 85114,
          "sample_id": "1f1a0cf61233ed36a5867777f0de217fa3b0481dfa6bfd86eb0b4e5bde7b7cc5",
          "token_shape": [
            1,
            90
          ],
          "tokens_npy_sha256": "a5bc6f3845fea9c38f8d140eb0e9c5b64fa39b8085310e18c16cb18a99b53993",
          "tokens_payload_sha256": "7606a9d0ee89caafc2de549e7c4c6b37037e1e0ff427777d7fab1ef861d7936e",
          "downloadedAtUnix": 1791562827.7655437,
          "shardBytes": 499434991,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 3.56,
              "num_samples_24k": 85440,
              "stored_prompt_tokens": 89,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/iptv/shard-af70f365f33b3f243dd80c19cd7dadba.npz",
          "shard_sha256": "ea5e30ce8f4dde482a1c974410e5e42e31325daeec24056db05ca236adefc132",
          "record_index": 25758,
          "sample_id": "1f1a0cf61233ed36a5867777f0de217fa3b0481dfa6bfd86eb0b4e5bde7b7cc5",
          "token_shape": [
            10,
            78
          ],
          "tokens_npy_sha256": "3d57ddb5054e46e9d6a5c61fdd0fd375a202739b0a55e22c7ad1c957584872be",
          "tokens_payload_sha256": "8f933782ebe1b05c046d7f4664e3729837d57a4880100234fb63bd63452cb7ec",
          "downloadedAtUnix": 1791562815.44467,
          "shardBytes": 497374254,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 158054,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 86400,
          "duration": 3.6,
          "sha256": "ce8bb45c04447430fec1d50361cdaa1e88335157c1caf642c5b1a9df45426645",
          "pcmSha256": "00d14fd1799f1058dca3991ef658d60e78756fc07609a1e74fb27e066b6c0fe8",
          "peak": 0.43516361713409424,
          "rms": 0.0585384290640373,
          "durationDeltaSeconds": 0.016000000000000014,
          "url": "audio/iptv-02.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 86400,
          "duration": 3.6,
          "sha256": "a4f124ba292a612ba8c6655f2b9dbdc83d96a28aa1c24abb8684db91bd46835c",
          "pcmSha256": "9af315d417a429d6be4c079f720035593c619e31682ecf956f1425fa9a3ebe31",
          "peak": 0.5834037065505981,
          "rms": 0.06606538703094184,
          "durationDeltaSeconds": 0.016000000000000014,
          "url": "audio/iptv-02.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 159744,
          "duration": 3.622312925170068,
          "sha256": "59e01013414ffdd2c2a93ccfb7107dbf145d70a65b095f469c566f9c81d47722",
          "pcmSha256": "ee10775f550ac950e38ede885a37bd1f433d5714a3ae34f9222205ecbfa3d408",
          "peak": 0.9985662698745728,
          "rms": 0.13378594092681992,
          "durationDeltaSeconds": 0.03831292517006801,
          "url": "audio/iptv-02.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 172032,
          "duration": 3.584,
          "sha256": "3fb15c7f662a9f2e435b4c4eeb12191a96fb7ad15aa881a82a24d665b0befbc5",
          "pcmSha256": "eebc0fe5435aadc050b34d6b245d02b84d7bb21015f796fb2c51500485969f39",
          "peak": 0.4827415347099304,
          "rms": 0.0641359780201115,
          "url": "audio/iptv-02.original.wav"
        }
      }
    },
    {
      "id": "iptv-03",
      "corpus": "iptv",
      "run": "english-35kh-20261006",
      "jobId": "1f4f8567-a7e3-54e4-a7ea-a0511e02ff2b",
      "sampleId": "a24b1c8da99e880d64146fd34261ebff4186f6a766eaf2559d96374bcd91ddc3",
      "selectionRank": 331566,
      "inputVariant": "original",
      "source": {
        "uri": "r2://iptvdata/CarChase.us/1786920884.aac.s555enc",
        "objectSha256": "06f8471186c76913517dc78058368fb040f69c21de1c3bf552a75c5aed4b23f1",
        "startSample": 16881504,
        "sampleRate": 48000,
        "numSamples": 508656,
        "pcmSha256": "a2466a94fa90cb9bbebde5036116b2b1dd07628142eadd283d4b08fdf3186eca",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/iptv/shard-b21b61979ada665f03d394472c685ed3.npz",
          "shard_sha256": "67a7dcec9926212eb46ecf949c643349a2a229cd32beca5a435c4ce2f1a2ad59",
          "record_index": 13398,
          "sample_id": "a24b1c8da99e880d64146fd34261ebff4186f6a766eaf2559d96374bcd91ddc3",
          "token_shape": [
            16,
            133
          ],
          "tokens_npy_sha256": "104e5e06d5746f1877da04383fc1d7ed5897b1ffb2cfa268deea9ac966b80955",
          "tokens_payload_sha256": "77a63e6cc344952ccfc77cfaaa93a86452bb82c5f92daa6310fe3388ea4e9e8a",
          "downloadedAtUnix": 1791562869.7227736,
          "shardBytes": 498035622,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1032,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/iptv/shard-66bc3fcee1c9f16fbc1a523dcdd69c96.npz",
          "shard_sha256": "23ee7ebc01447f10d33365f0ad2384a5fcf42f97d6f7ffd6b17ca0bd5f2954f7",
          "record_index": 41638,
          "sample_id": "a24b1c8da99e880d64146fd34261ebff4186f6a766eaf2559d96374bcd91ddc3",
          "token_shape": [
            1,
            265
          ],
          "tokens_npy_sha256": "0cd2be05930a14cf702e1f4c4056dbe0919c5c93b3c11d2f7481f202fb645020",
          "tokens_payload_sha256": "deef396d5a48926ccfa327072f22f52988834415fda0b547b8a6c88b2016627a",
          "downloadedAtUnix": 1791562897.1920218,
          "shardBytes": 497454989,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/iptv/shard-d866510a1753271178731cba090724e5.npz",
          "shard_sha256": "6d1a8621c9a3b37443cdba480fafdadfe7f2a1d9c4556ecb77359b9e7dd0e6bf",
          "record_index": 12283,
          "sample_id": "a24b1c8da99e880d64146fd34261ebff4186f6a766eaf2559d96374bcd91ddc3",
          "token_shape": [
            10,
            229
          ],
          "tokens_npy_sha256": "b3502376bba5011cd57eae11d15decf738d4b2ef5770c1aba946cc8fff9d61fd",
          "tokens_payload_sha256": "7291bfdc45b6e36c6867b39c456017589b87d382b4fa96c875f5ab1e73909593",
          "downloadedAtUnix": 1791562890.9348047,
          "shardBytes": 498425580,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 467327,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 255360,
          "duration": 10.64,
          "sha256": "d5f32453340277f940213dcac0efc7a04e62490b3cf5864f62c32689f7ee5e8b",
          "pcmSha256": "9df0a183abdbeea8d5a28c22949cea4d5e70345902ac94a61831a4cbb611cb72",
          "peak": 0.7644615769386292,
          "rms": 0.05995601366477567,
          "durationDeltaSeconds": 0.04300000000000104,
          "url": "audio/iptv-03.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 254400,
          "duration": 10.6,
          "sha256": "f04fd040868c78477b03bf7db1d9b30831ecac9565323e9f7e1b912ae858e681",
          "pcmSha256": "8d05ab2664bf2ecc7bdfde313136b4fbe826cf02dfd03ce7c723387656c18a5c",
          "peak": 0.8351220488548279,
          "rms": 0.07221273630769272,
          "durationDeltaSeconds": 0.0030000000000001137,
          "url": "audio/iptv-03.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 468992,
          "duration": 10.634739229024943,
          "sha256": "a4b5f62a46110ac75c5fec582c13c34c3e5c177bba94005b1aab139d885577f6",
          "pcmSha256": "1e3c7668c7c4652bf51b37cb9e7580b1844c1d99b82e38dced8aacdab2ba271a",
          "peak": 0.9988076090812683,
          "rms": 0.14574351644462352,
          "durationDeltaSeconds": 0.0377392290249432,
          "url": "audio/iptv-03.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 508656,
          "duration": 10.597,
          "sha256": "353c1d8806c69265860ad08253c3875e198366ee0b039fffbbd43fe707fb3949",
          "pcmSha256": "a2466a94fa90cb9bbebde5036116b2b1dd07628142eadd283d4b08fdf3186eca",
          "peak": 0.6517246961593628,
          "rms": 0.06193873972462957,
          "url": "audio/iptv-03.original.wav"
        }
      }
    },
    {
      "id": "iptv-04",
      "corpus": "iptv",
      "run": "english-35kh-20261006",
      "jobId": "824b6897-9817-5cd3-b37f-404e67f93866",
      "sampleId": "ef81bf190e451cfeb756f8af5851587509150d81ea03ab335114fdaa0474d627",
      "selectionRank": 1395086,
      "inputVariant": "original",
      "source": {
        "uri": "r2://iptvdata/CityNewsEdmonton.ca/1787267270.aac.s555enc",
        "objectSha256": "201067655a0a290c677a291141cd17e9ef938ec8d40106ac0cdf9a77da5ed09c",
        "startSample": 18000816,
        "sampleRate": 48000,
        "numSamples": 611472,
        "pcmSha256": "f15a80708844e728cfb3e563ce22a18428d37ea75d0b452dad9dba8c25cbe94e",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/iptv/shard-bac9d51406b48d62ef42674dabb988c7.npz",
          "shard_sha256": "aeaad1488f11a550b2060bdc37461e801d8799828d61e772c3a76a5d4e4843bf",
          "record_index": 50338,
          "sample_id": "ef81bf190e451cfeb756f8af5851587509150d81ea03ab335114fdaa0474d627",
          "token_shape": [
            16,
            160
          ],
          "tokens_npy_sha256": "de6f845f10af434242f42cea6cd38729290f4b8d58c7f32ca5cd33cd5c16241b",
          "tokens_payload_sha256": "35c26f0292f45ebcbfc8606c98e9bb47325d2dfdf3a5951976bd637138fd341f",
          "downloadedAtUnix": 1791562894.8340654,
          "shardBytes": 499926720,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1464,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/iptv/shard-07482a9eefdc717f481dd8b5fd9fc8a0.npz",
          "shard_sha256": "6cf3a0437c0c8acc6a7266bb6988da80536b4d1bf2967600bf4b475728dd3cb3",
          "record_index": 177063,
          "sample_id": "ef81bf190e451cfeb756f8af5851587509150d81ea03ab335114fdaa0474d627",
          "token_shape": [
            1,
            319
          ],
          "tokens_npy_sha256": "7a59ef920555044f504004389b352b773e43e8dce559dbf35d0be5881a4723d9",
          "tokens_payload_sha256": "e7deeb09467d9b968f09542a2e8c8f32a6629669eb3baf2b4f228cedc4505dc4",
          "downloadedAtUnix": 1791562899.9047287,
          "shardBytes": 497602965,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/iptv/shard-4a10acc18dbc3cb5b4a3316320da3fd0.npz",
          "shard_sha256": "d14df8ceeaa8129469fdd52442f6c5bcc027d453c720a89971249245e12a365d",
          "record_index": 49522,
          "sample_id": "ef81bf190e451cfeb756f8af5851587509150d81ea03ab335114fdaa0474d627",
          "token_shape": [
            10,
            275
          ],
          "tokens_npy_sha256": "1f229c825ec721458854d82cdf25cf34c6345f66e478f8fd984292db060dfc35",
          "tokens_payload_sha256": "95dc4517fcc00407ba1f95c33beb2857d852ab97fd1b3fd0f08d32ff3c53a64b",
          "downloadedAtUnix": 1791562889.7051694,
          "shardBytes": 499665243,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 561789,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 307200,
          "duration": 12.8,
          "sha256": "66ecd8b5fecb5a8707714f5e41c986d44da6922f35f001d6ea9c5459b06549a4",
          "pcmSha256": "184500f5b84817bff1971a7f8ab6bc72c942e5291afd36803ccd4e36353bd17a",
          "peak": 0.6706425547599792,
          "rms": 0.09019776323805175,
          "durationDeltaSeconds": 0.06099999999999994,
          "url": "audio/iptv-04.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 306240,
          "duration": 12.76,
          "sha256": "6c3d7db392b30ca7fb96322387d1be7c953beb5eb2be2378e03e079b8cc1ea52",
          "pcmSha256": "1a5e9347db1f42701c6d3315afc39e42cf0143f3d254a4bd7cafffab81cf87f3",
          "peak": 0.7245485782623291,
          "rms": 0.10855638508388624,
          "durationDeltaSeconds": 0.02099999999999902,
          "url": "audio/iptv-04.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 563200,
          "duration": 12.770975056689343,
          "sha256": "40ef7b845f40622299a0b6c0d1793f4ab435c2d72aa83f90af6e4003b2acfffd",
          "pcmSha256": "6091399559bb23fa7e8cd474bcfb618e05fba301c75ebd54b0828df862ad2ded",
          "peak": 0.999219536781311,
          "rms": 0.14824788785631465,
          "durationDeltaSeconds": 0.031975056689342196,
          "url": "audio/iptv-04.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 611472,
          "duration": 12.739,
          "sha256": "1b534449622ed4a2583e529e44fe3d0064a0983889b67651897a9af2842cfbf8",
          "pcmSha256": "f15a80708844e728cfb3e563ce22a18428d37ea75d0b452dad9dba8c25cbe94e",
          "peak": 0.8739739060401917,
          "rms": 0.0940355699419974,
          "url": "audio/iptv-04.original.wav"
        }
      }
    },
    {
      "id": "iptv-05",
      "corpus": "iptv",
      "run": "english-35kh-20261006",
      "jobId": "11ac2372-db1d-53f8-8d00-cdf2fca4720d",
      "sampleId": "6ad5d41d99ed6792864a3e49b100c0bfc39a736b056a6cb40be6d4b0c851d4c0",
      "selectionRank": 185962,
      "inputVariant": "original",
      "source": {
        "uri": "r2://iptvdata/CityNewsVancouver.ca/1786156359.aac.s555enc",
        "objectSha256": "8f724e364b02250a677dd00b5c3ca2f9805500836c5ff43ae225712544d7116a",
        "startSample": 0,
        "sampleRate": 48000,
        "numSamples": 960000,
        "pcmSha256": "9238a65b1a7892cd3b99219dda70ecf3bec217d9f0bf9bdf9a37648be9dff6b3",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/iptv/shard-aad9f43b4dd0981d9aef1e892411ac6d.npz",
          "shard_sha256": "3edbea39fc172032b7b623d7b4f323b08ecada1567505e13b28e5ba39bdd868e",
          "record_index": 7088,
          "sample_id": "6ad5d41d99ed6792864a3e49b100c0bfc39a736b056a6cb40be6d4b0c851d4c0",
          "token_shape": [
            16,
            250
          ],
          "tokens_npy_sha256": "10eb6dc567ccc4ee98096b1fed2a37af9a45a37063ea13755d0afd2c06053fea",
          "tokens_payload_sha256": "1be3eea0dde7e5ec62e97f7cef1dae6d13e1b15ec4d1c67a1715879adcbd313e",
          "downloadedAtUnix": 1791562917.2519338,
          "shardBytes": 498707545,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 0,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/iptv/shard-07482a9eefdc717f481dd8b5fd9fc8a0.npz",
          "shard_sha256": "6cf3a0437c0c8acc6a7266bb6988da80536b4d1bf2967600bf4b475728dd3cb3",
          "record_index": 23782,
          "sample_id": "6ad5d41d99ed6792864a3e49b100c0bfc39a736b056a6cb40be6d4b0c851d4c0",
          "token_shape": [
            1,
            500
          ],
          "tokens_npy_sha256": "a5c4ef463a5e4a70ab2555d08dbef88b3acc8816c5e37f075ff99241881e1009",
          "tokens_payload_sha256": "b1ad5425f296d04649eb65f74bb102f6a9fc78d59f9d342a229f773ea7656805",
          "downloadedAtUnix": 1791562899.9047287,
          "shardBytes": 497602965,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/iptv/shard-4a10acc18dbc3cb5b4a3316320da3fd0.npz",
          "shard_sha256": "d14df8ceeaa8129469fdd52442f6c5bcc027d453c720a89971249245e12a365d",
          "record_index": 6363,
          "sample_id": "6ad5d41d99ed6792864a3e49b100c0bfc39a736b056a6cb40be6d4b0c851d4c0",
          "token_shape": [
            10,
            431
          ],
          "tokens_npy_sha256": "589728b38de7af6693a9ea3fd89b1e766e55459b07da51cc0679b3742f6adef3",
          "tokens_payload_sha256": "b9057a9f148bbc1e8c74906c9258f0a6a9602b937835cc7260e0ca553bd2c602",
          "downloadedAtUnix": 1791562889.7051694,
          "shardBytes": 499665243,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 882000,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 480000,
          "duration": 20.0,
          "sha256": "1169b525c9ac9878a0b861e7ca42a041a0c2e5471b08979aa56fa7a2bd6e5e3f",
          "pcmSha256": "57a9fd64cd630220d23a4728d037f088506e17973bcba23896b48e8ef5a328c0",
          "peak": 0.3910382390022278,
          "rms": 0.05773431480745225,
          "durationDeltaSeconds": 0.0,
          "url": "audio/iptv-05.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 480000,
          "duration": 20.0,
          "sha256": "5950da7cfbb1dd188d8867302f24a26ed41ab8324da6939ed7b614bdd6f8cfbc",
          "pcmSha256": "17b841deebbeb95e065f88a8a56a352a2b06e7b10915cc30a0b566ac1c058002",
          "peak": 0.6634292602539062,
          "rms": 0.06817580278762923,
          "durationDeltaSeconds": 0.0,
          "url": "audio/iptv-05.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 882688,
          "duration": 20.01560090702948,
          "sha256": "17b217026b2d9229fdbe86450be638bca83adc9963f3c8313501b39bb134b19c",
          "pcmSha256": "1f445826fd9fc796e3289912b647ae6bce8ddbadcff7613c585c22ba9ae0b406",
          "peak": 0.9995764493942261,
          "rms": 0.16167439773716313,
          "durationDeltaSeconds": 0.015600907029480027,
          "url": "audio/iptv-05.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 960000,
          "duration": 20.0,
          "sha256": "446ce2a5411d174a3c0103fa84736593b47bf14cc9579a30f67549ea70a3b521",
          "pcmSha256": "9238a65b1a7892cd3b99219dda70ecf3bec217d9f0bf9bdf9a37648be9dff6b3",
          "peak": 0.38752731680870056,
          "rms": 0.06013861872868637,
          "url": "audio/iptv-05.original.wav"
        }
      }
    },
    {
      "id": "podcast-01",
      "corpus": "podcast",
      "run": "english-35kh-20261006",
      "jobId": "b41ba354-595a-433e-889b-3db8b18c15c8",
      "sampleId": "b25b5b62183a0f3ad7f38c1f1826eeb8e679dc946ae43431010dc63b3150ff2a",
      "selectionRank": 3805698,
      "inputVariant": "original",
      "source": {
        "uri": "r2://episodess/U/9b6f35fe7300da9fc0ae6c77fbf9d125/8ef5f5fc6bc7e18d6ff9d4af30115d95.mp3.enc",
        "objectSha256": "40c06bc6598a3e435e8beb56d2a39bd63e3bccd37f8330df8139dca5c5ea020a",
        "startSample": 61216800,
        "sampleRate": 48000,
        "numSamples": 343200,
        "pcmSha256": "dd74cfbe6c37705b4cc760685e02c0a62d3a60ce543d89419c0216ab4e637ba1",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/podcast/shard-daeb02bdf014ad699db0724abcc308c6.npz",
          "shard_sha256": "d306e67963228430e2f8622491110f8ddfca572106c15c1797b8f11af46751eb",
          "record_index": 75306,
          "sample_id": "b25b5b62183a0f3ad7f38c1f1826eeb8e679dc946ae43431010dc63b3150ff2a",
          "token_shape": [
            16,
            90
          ],
          "tokens_npy_sha256": "86d8a2f416633c703de28a5506b18e7417180d2186406d8dced8e2ac8c64adfd",
          "tokens_payload_sha256": "97af847def5d8d9af4d42c92a87689cf99b62ed81134d4b8779c197a7946f00f",
          "downloadedAtUnix": 1791562937.3579345,
          "shardBytes": 490284680,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1200,
            "samples_at_native_clamp": 16
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/podcast/shard-4429bb06d019b6dea8890dcb33a2d660.npz",
          "shard_sha256": "7c5ff1829bc93d9ec2813621b675e2e8ac7cd37788b0d0fa86b4b2b520f8c465",
          "record_index": 336747,
          "sample_id": "b25b5b62183a0f3ad7f38c1f1826eeb8e679dc946ae43431010dc63b3150ff2a",
          "token_shape": [
            1,
            179
          ],
          "tokens_npy_sha256": "0d15fde2d12c4931c6cc5a5e63c86da3b151c3c5dbb85035137aad5a9af9ab76",
          "tokens_payload_sha256": "aac1b87262937e826ed833535f132bc2e8878aaf6cad132d56aba069858e5370",
          "downloadedAtUnix": 1791562930.2476423,
          "shardBytes": 499428001,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 7.12,
              "num_samples_24k": 170880,
              "stored_prompt_tokens": 178,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/podcast/shard-14c3c288db78ae79d4994bdb3aef07f6.npz",
          "shard_sha256": "aed3d4750010917cea456709ee1b00f66d447ebcc14a8183b575b76a0e3c7b53",
          "record_index": 74198,
          "sample_id": "b25b5b62183a0f3ad7f38c1f1826eeb8e679dc946ae43431010dc63b3150ff2a",
          "token_shape": [
            10,
            154
          ],
          "tokens_npy_sha256": "ea2b66b53ae404d714479b997adcd7f67b16c9340e679d9df1a99ae879344307",
          "tokens_payload_sha256": "ce94b6e6a985351bb3383e91f83d6ea810aadbb1115e0935697425b23f32401a",
          "downloadedAtUnix": 1791562930.2759442,
          "shardBytes": 492815889,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 315315,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 172800,
          "duration": 7.2,
          "sha256": "1b55eb7c5cb088628ed1f4f29e7224d1de7ec6d30cd7a16eaa40e63d889f795c",
          "pcmSha256": "a8aad48bbb8e483762b221e85b85b8b05191777b3cfddfab15a710ce83f4c03c",
          "peak": 1.0,
          "rms": 0.15362624755309645,
          "durationDeltaSeconds": 0.04999999999999982,
          "url": "audio/podcast-01.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 171840,
          "duration": 7.16,
          "sha256": "2e4ec9539f76872d5800161e41cfdbec6a9ee2aadf56c2178589eedf869accd7",
          "pcmSha256": "e4792cd5cc44e88c3db83217b7a3f1d6d3f4ebb5b3a0c7e6e459e40c715f56b0",
          "peak": 0.9900000095367432,
          "rms": 0.16555679779050136,
          "durationDeltaSeconds": 0.009999999999999787,
          "url": "audio/podcast-01.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 315392,
          "duration": 7.151746031746032,
          "sha256": "40c3af6db56edae3b3c0d8e376a391d252760ff76d6a5bb8ab26e6cc79a760f6",
          "pcmSha256": "3b2b3dbe5060ffea36ae70f7ee60c6518d212611c514c8c6ee52212f282c2d95",
          "peak": 0.999157190322876,
          "rms": 0.14516839278528146,
          "durationDeltaSeconds": 0.0017460317460313846,
          "url": "audio/podcast-01.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 343200,
          "duration": 7.15,
          "sha256": "0b5d2d1dc17836d7da4bf1e60c4630396c258ecafb45d06a5f683f777d8ac292",
          "pcmSha256": "dd74cfbe6c37705b4cc760685e02c0a62d3a60ce543d89419c0216ab4e637ba1",
          "peak": 1.0531747341156006,
          "rms": 0.161160081838404,
          "url": "audio/podcast-01.original.wav"
        }
      }
    },
    {
      "id": "podcast-02",
      "corpus": "podcast",
      "run": "english-35kh-20261006",
      "jobId": "3348511a-ccd6-4098-9934-75c54b14b449",
      "sampleId": "318de86a6d262c11815171049818b81dc0940d332d35b0ff9cc9d4af7ccc4c4c",
      "selectionRank": 1076122,
      "inputVariant": "original",
      "source": {
        "uri": "r2://episodess/G/5f04de44b3b2ce0fd8d2c0f075f9ecb1/c9e638e85a92470d91f67b029fdb53e4.mp3.enc",
        "objectSha256": "8273175fa8e4724c009e00ea6a4bbaa847bb0e7d56d4b7afaf82a80b5cf55aeb",
        "startSample": 43693536,
        "sampleRate": 48000,
        "numSamples": 860448,
        "pcmSha256": "5b7618ab13dbc95016871c10d74961fdc77ea8578aee3dca18c4b3bfa07e30d9",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/podcast/shard-a1d1efa9b3b349fba4b5b6b59aeb70a2.npz",
          "shard_sha256": "6c7428044726bc169b750d575ad8e400f18f9c4a8de1247344b98e42954b418d",
          "record_index": 38217,
          "sample_id": "318de86a6d262c11815171049818b81dc0940d332d35b0ff9cc9d4af7ccc4c4c",
          "token_shape": [
            16,
            225
          ],
          "tokens_npy_sha256": "e28e93eaec179abc4f64e32a77a96c56d4db6c7a3595e02bbe0e6bf2d4ce2541",
          "tokens_payload_sha256": "8a3f6adeccf74f426f8596e20068423be14a7d6e7dd1a66c329a135a1e0e3b68",
          "downloadedAtUnix": 1791562939.1643908,
          "shardBytes": 496849375,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1776,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/podcast/shard-c18e4ba61e3dfae7858711cf3a1a3bc0.npz",
          "shard_sha256": "53789a69a60337fde225f36ed3f693ef49a6738f2e6c5b7afb9dd0c970abb6ca",
          "record_index": 92230,
          "sample_id": "318de86a6d262c11815171049818b81dc0940d332d35b0ff9cc9d4af7ccc4c4c",
          "token_shape": [
            1,
            448
          ],
          "tokens_npy_sha256": "3fb24a7e0fe4beac6a2afc51c57b84e75d7118c85de9668925b90f0cb1de0792",
          "tokens_payload_sha256": "51f53146781d9d6782eee994f9db9a728cafbd7078afcda439400973e1411a19",
          "downloadedAtUnix": 1791562953.478667,
          "shardBytes": 492589751,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/podcast/shard-868f7182f3183f413734c68a2a54d8fd.npz",
          "shard_sha256": "d2d6c31ae6c0f3b0e4e5fa760e6850a5ea94dbb24bcbe815b0b88d79eb46d12d",
          "record_index": 50494,
          "sample_id": "318de86a6d262c11815171049818b81dc0940d332d35b0ff9cc9d4af7ccc4c4c",
          "token_shape": [
            10,
            387
          ],
          "tokens_npy_sha256": "a477de32e413b42a8f39c053ff10f17c6dc8e3f8b3e2cb82028eb0e28d6f35b1",
          "tokens_payload_sha256": "c5ce96c2e79fa89d9bd7be172536563e5e969603a4d9b7d93ea91e6ca92becf8",
          "downloadedAtUnix": 1791562979.1798484,
          "shardBytes": 497387787,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 790536,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 432000,
          "duration": 18.0,
          "sha256": "c762fa0366d1ce71f042c5d4bb1928efcde7c7e9f34694ca6f54025290c4f4a0",
          "pcmSha256": "1d9eb72f393ad86633a12f0ae807d946f23ec4ba5a3a97ea1bc0bc2232110da4",
          "peak": 0.4579137861728668,
          "rms": 0.04723014721740234,
          "durationDeltaSeconds": 0.07400000000000162,
          "url": "audio/podcast-02.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 430080,
          "duration": 17.92,
          "sha256": "4c99986105b72a22216f04950f82807f833e55419ea1039faaa1afcf7253b1ec",
          "pcmSha256": "d9aa162ebfaac653163789b5278be9ecccfcac2ff3891517ffdb9075ea46a0f4",
          "peak": 0.521512508392334,
          "rms": 0.055823174579122,
          "durationDeltaSeconds": -0.005999999999996675,
          "url": "audio/podcast-02.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 792576,
          "duration": 17.972244897959182,
          "sha256": "7ad21f3410433e36383cf202f146099471a1f10f786242d088a203c07751b24b",
          "pcmSha256": "6908e996176791baa39f2ca875f7e21eaa7fa084289b74a4ea7ce29d17f8bbf5",
          "peak": 0.9994637966156006,
          "rms": 0.1375027350332897,
          "durationDeltaSeconds": 0.04624489795918407,
          "url": "audio/podcast-02.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 860448,
          "duration": 17.926,
          "sha256": "b3e75badc09ab18bf92285e7127ff409de236c405e04e8c79d30238addafd763",
          "pcmSha256": "5b7618ab13dbc95016871c10d74961fdc77ea8578aee3dca18c4b3bfa07e30d9",
          "peak": 0.5138111114501953,
          "rms": 0.048688004150661206,
          "url": "audio/podcast-02.original.wav"
        }
      }
    },
    {
      "id": "podcast-03",
      "corpus": "podcast",
      "run": "english-35kh-20261006",
      "jobId": "c2d51ee2-3833-4393-9423-6dc8030113e7",
      "sampleId": "cab3e69bba57fdd9669a098ec866748f576df135999ad85badde4dee20c1b9b2",
      "selectionRank": 4115633,
      "inputVariant": "original",
      "source": {
        "uri": "r2://episodess/M/68d3f669c2778c0475e461ba893d2e7b/bc823c62b2f2bae448ed0ef8a522df91.mp3.enc",
        "objectSha256": "7fea245a1039cfb9e398cb53a95569f4040e13e37039902718b7b35233f32e50",
        "startSample": 87889536,
        "sampleRate": 48000,
        "numSamples": 910464,
        "pcmSha256": "86e6cba2785b3176f98fcc589959fe19f6133071dd165a2d1ea7fef5b5274499",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/podcast/shard-9761a4042704e8dc7e317a8738f5d9cd.npz",
          "shard_sha256": "3711e56ab8d1cc865092439a485c30559b30cc521b16ffc604b50b87ba830f8c",
          "record_index": 73296,
          "sample_id": "cab3e69bba57fdd9669a098ec866748f576df135999ad85badde4dee20c1b9b2",
          "token_shape": [
            16,
            238
          ],
          "tokens_npy_sha256": "26319094fab653e1952d5585999f96a7243e42f204bfe3377557a57c0b3cd225",
          "tokens_payload_sha256": "8c1698af2830a20ad50d2191282f64bc9b82c5bae8977bab1ee2bf92285a7c4e",
          "downloadedAtUnix": 1791562979.8859053,
          "shardBytes": 497339177,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1728,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/podcast/shard-c18e4ba61e3dfae7858711cf3a1a3bc0.npz",
          "shard_sha256": "53789a69a60337fde225f36ed3f693ef49a6738f2e6c5b7afb9dd0c970abb6ca",
          "record_index": 368620,
          "sample_id": "cab3e69bba57fdd9669a098ec866748f576df135999ad85badde4dee20c1b9b2",
          "token_shape": [
            1,
            474
          ],
          "tokens_npy_sha256": "911942c8490ac4bdaf79ff0d312d91d77c73238c3d0f009d96c34ffad07c37d6",
          "tokens_payload_sha256": "98ed89c39cad22f4e4a072363319821cc8756b4e0ae9809d842dec95af5fe01f",
          "downloadedAtUnix": 1791562953.478667,
          "shardBytes": 492589751,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/podcast/shard-fd2277d6c4b717bfb2118aee752f7c1b.npz",
          "shard_sha256": "beadda46ddf66469321966281eea493fbc879a5a14d64c2ed69108685ec2d4b3",
          "record_index": 94426,
          "sample_id": "cab3e69bba57fdd9669a098ec866748f576df135999ad85badde4dee20c1b9b2",
          "token_shape": [
            10,
            409
          ],
          "tokens_npy_sha256": "3f504bfa303443f6a2997d7ff05246565f130ba4b0ac63ae88d864d277ad29d9",
          "tokens_payload_sha256": "10264297ee61b1ed29fc50075e3751e19b55a29e872a88a7af06531283ccce64",
          "downloadedAtUnix": 1791562997.4724975,
          "shardBytes": 496960191,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 836488,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 456960,
          "duration": 19.04,
          "sha256": "7f9d9951df1492269f5e30e719e917495bc2c0bb7e9ebbe49182088b3d18b105",
          "pcmSha256": "2efa187a891e3a66a97b78419077aafe8c4838385c97f525daaa00c4d1e7bdc6",
          "peak": 0.7242071032524109,
          "rms": 0.1082447809927717,
          "durationDeltaSeconds": 0.07199999999999918,
          "url": "audio/podcast-03.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 455040,
          "duration": 18.96,
          "sha256": "88051264dc155212c97b6be6844357b50829ddb45cb6ef15669b0a0932b2386c",
          "pcmSha256": "e1aefbcb0c3859a0af17b7ddee661ec61bc99f9e26007e2ec0c9943960ea9562",
          "peak": 0.9622204303741455,
          "rms": 0.11629132931249139,
          "durationDeltaSeconds": -0.007999999999999119,
          "url": "audio/podcast-03.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 837632,
          "duration": 18.99392290249433,
          "sha256": "8088979b371ba5c003028c2a054045ac78764dbdbb20d7399711fc954e4521a3",
          "pcmSha256": "d25c9e7b83de9673154d74b8bad71c8fe1f3052b6d7327f2805b979f0b321068",
          "peak": 0.9992010593414307,
          "rms": 0.1528851764877331,
          "durationDeltaSeconds": 0.02592290249432949,
          "url": "audio/podcast-03.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 910464,
          "duration": 18.968,
          "sha256": "e9aae4765a815fc151b86324b37f8017043de2f7e1c99524254c3569ae867438",
          "pcmSha256": "86e6cba2785b3176f98fcc589959fe19f6133071dd165a2d1ea7fef5b5274499",
          "peak": 0.8051893711090088,
          "rms": 0.10979974023589009,
          "url": "audio/podcast-03.original.wav"
        }
      }
    },
    {
      "id": "podcast-04",
      "corpus": "podcast",
      "run": "english-35kh-20261006",
      "jobId": "47dc1030-2131-4da4-8a8a-6637b1001226",
      "sampleId": "cc20c1edf673d8a47f4305dd8d8c341153f30c159c8e419a79e7704ec81a3cfc",
      "selectionRank": 1515487,
      "inputVariant": "original",
      "source": {
        "uri": "r2://episodess/O/410bff4716f0374ca51e23b616979d84/272be5be49cb605de8647296f4be7cf4.mp3.enc",
        "objectSha256": "1e31bf81c0ccc3368568432506bd99f1f10cca48ca4cbdfac8b9c0b7ae459de8",
        "startSample": 97075920,
        "sampleRate": 48000,
        "numSamples": 423456,
        "pcmSha256": "a678206d1be73ad000ab0afb3aa7f8d2f44d54260cc79bfd719f9281c2566888",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/podcast/shard-a67fec5351d8952f8f14a1832ca690bf.npz",
          "shard_sha256": "5614ba2715c44ec8164d22def1526e75b3289ac1e7779b6a7ee9aa56200fe147",
          "record_index": 31809,
          "sample_id": "cc20c1edf673d8a47f4305dd8d8c341153f30c159c8e419a79e7704ec81a3cfc",
          "token_shape": [
            16,
            111
          ],
          "tokens_npy_sha256": "96b291f8a8ab2676e040ff7c505138f4b555584560dfbd9a63e5c5e57ce76e97",
          "tokens_payload_sha256": "628bbf1674683d7972858900bbd36a7b3d9191462f85e0c199abc721bfb0087f",
          "downloadedAtUnix": 1791563008.990067,
          "shardBytes": 497475408,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1392,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/podcast/shard-bf2f3243c0fd8d293f5cc5cd6f64c6db.npz",
          "shard_sha256": "94053d61d93f6429bff996b3a89d658b5e7a90b518d887c6ec25eb6f6bc39bdc",
          "record_index": 128837,
          "sample_id": "cc20c1edf673d8a47f4305dd8d8c341153f30c159c8e419a79e7704ec81a3cfc",
          "token_shape": [
            1,
            221
          ],
          "tokens_npy_sha256": "0837acb333b7e26d231069054a3aefcf67fc1faa72a373ef85a78cca964d3e6e",
          "tokens_payload_sha256": "356e12bbf9d25049fe7efe4cd0e9ea4a9fda258b39c514b23b9497da6ffa913d",
          "downloadedAtUnix": 1791562994.285798,
          "shardBytes": 498934601,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 8.8,
              "num_samples_24k": 211200,
              "stored_prompt_tokens": 220,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/podcast/shard-2cfa158b320c5412746eb13298a1f94d.npz",
          "shard_sha256": "b631002d5b97e61e489e6e66e12351f8665257d8301a344c81ff2aa2cedf2982",
          "record_index": 22124,
          "sample_id": "cc20c1edf673d8a47f4305dd8d8c341153f30c159c8e419a79e7704ec81a3cfc",
          "token_shape": [
            10,
            190
          ],
          "tokens_npy_sha256": "b2b104201cd36204845193e418e02fa85eb8e943042983c273f14f93859d224b",
          "tokens_payload_sha256": "7ba3443ceb69a7915a334fc899e4ba5c7e6a4d12dfee55ccea93ccc39916a452",
          "downloadedAtUnix": 1791563013.4720287,
          "shardBytes": 497192477,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 389050,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 213120,
          "duration": 8.88,
          "sha256": "6a51ab9eb6f8243859cc6ef06faf09963333facb87558a25aafce57cb3aff018",
          "pcmSha256": "4e3596378d558a9aea704d00f2b63a1fab23f0714b42246dc92e1cf7c30cb064",
          "peak": 0.5235210657119751,
          "rms": 0.047425882948449585,
          "durationDeltaSeconds": 0.058000000000001606,
          "url": "audio/podcast-04.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 212160,
          "duration": 8.84,
          "sha256": "c5cb3a4599f17238bb22ead97299e246f0e9b6a8e405a94423e8e0bc2967d2a1",
          "pcmSha256": "3cbfe97b29d5d5d282edd863b1cec434fc087f0f25e9699bead74477e0ae83ad",
          "peak": 0.5425654053688049,
          "rms": 0.054801875778670156,
          "durationDeltaSeconds": 0.018000000000000682,
          "url": "audio/podcast-04.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 389120,
          "duration": 8.823582766439909,
          "sha256": "5395157e2df9b8441c20c1ae3dfb33dc46cae9bf6a56e3f87020158959cbad8d",
          "pcmSha256": "17d6d50d9cdb266a9780652bde521ec6deaa92f0b347f963db136113855673ad",
          "peak": 0.9993485808372498,
          "rms": 0.13544174935127262,
          "durationDeltaSeconds": 0.001582766439909733,
          "url": "audio/podcast-04.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 423456,
          "duration": 8.822,
          "sha256": "b61edaba6999ec1dcae98d8259420c26f0fa96cd861bdf333acc8766f2e49b11",
          "pcmSha256": "a678206d1be73ad000ab0afb3aa7f8d2f44d54260cc79bfd719f9281c2566888",
          "peak": 0.544930100440979,
          "rms": 0.05034044320476838,
          "url": "audio/podcast-04.original.wav"
        }
      }
    },
    {
      "id": "podcast-05",
      "corpus": "podcast",
      "run": "english-35kh-20261006",
      "jobId": "45c719b1-ae57-4657-935d-26b59ad40f01",
      "sampleId": "c64905228b92b80c8bc31b66f96d5af5973bc30063e26059d3f0e6d618840e8c",
      "selectionRank": 1468430,
      "inputVariant": "original",
      "source": {
        "uri": "r2://episodess/B/2a6d01fc1a330938fdc536879dc81bf3/f15e51c52f216d273d6fae6123ee63cd.mp3.enc",
        "objectSha256": "2b2f00cc1d42bbb8488ef6f15a8583367c7044b3ed7a83b9cf539e24a964cf2f",
        "startSample": 2950032,
        "sampleRate": 48000,
        "numSamples": 756432,
        "pcmSha256": "2ab64f1668212f4ef0312aa42991357d764a169010c559ef4ff707322654160a",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/qwen/podcast/shard-fc8ee84c66fa12744dcaf983294a335d.npz",
          "shard_sha256": "b2e5e3ad201d6cdc01283b44309fe1fd4887ad3b0ab276a48908ee05281be269",
          "record_index": 29222,
          "sample_id": "c64905228b92b80c8bc31b66f96d5af5973bc30063e26059d3f0e6d618840e8c",
          "token_shape": [
            16,
            197
          ],
          "tokens_npy_sha256": "4702949f1c7d5bc782d24f0f9159100cff3e6e0890579cf7da385c1cc7a97379",
          "tokens_payload_sha256": "32a42615658548352abb750f6df2f9c51a27c23cc4cbceba15b0a5f7da13ea28",
          "downloadedAtUnix": 1791563037.697802,
          "shardBytes": 497180566,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 24,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/chatterbox/podcast/shard-8863601549e597f23d92caf32957f623.npz",
          "shard_sha256": "f9acfc86c5094d8a1a71f5495bf380331f7565b3d50433d861bc54911f11dbc0",
          "record_index": 126490,
          "sample_id": "c64905228b92b80c8bc31b66f96d5af5973bc30063e26059d3f0e6d618840e8c",
          "token_shape": [
            1,
            394
          ],
          "tokens_npy_sha256": "90247bcb41c2787e5d9d0e590baf74ee754d4b22d5716e533a71edc74ba8b7df",
          "tokens_payload_sha256": "01293d77100821472d8b34b7d327085390624856e7e95248e4a275412db00a6a",
          "downloadedAtUnix": 1791563046.1034818,
          "shardBytes": 499605171,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "english-35kh-20261006/codecs/fish/podcast/shard-6fbc0d5c2c7c4f4c4721503a3a25e8e9.npz",
          "shard_sha256": "b0cd7e28a3461762cc4e0f76f98490007b0c753f2a17f2d023b0a492490cf9fe",
          "record_index": 26618,
          "sample_id": "c64905228b92b80c8bc31b66f96d5af5973bc30063e26059d3f0e6d618840e8c",
          "token_shape": [
            10,
            340
          ],
          "tokens_npy_sha256": "f9999f4e039e07e961fd75b80a6e267b5a102be7b7447e388edcd718638f3f75",
          "tokens_payload_sha256": "a09f7fec435f053c5381d241d26ed1a75d66287bd9390e357dbc003c94208415",
          "downloadedAtUnix": 1791563061.739713,
          "shardBytes": 497459214,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fishdac",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 694971,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 378240,
          "duration": 15.76,
          "sha256": "f1a9ad72a5a11a345e6873e9e767c96f7608a72a333b96a557b04304835ff76c",
          "pcmSha256": "85e7af03a828a75569615e282e6cc6cd3853f65a23c20ea0cb3b3bce7f1c0ceb",
          "peak": 0.6846172213554382,
          "rms": 0.07545875599074849,
          "durationDeltaSeconds": 0.0009999999999994458,
          "url": "audio/podcast-05.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 378240,
          "duration": 15.76,
          "sha256": "35f88513056ad5cd3e30fec710dee6bed7d079be8fbe0b53103bcad705610c5f",
          "pcmSha256": "a9b1be5b77084d04cbef217326be0127bfd2f5b6894b205a5b462039e7b15763",
          "peak": 0.8011326789855957,
          "rms": 0.0855586530550151,
          "durationDeltaSeconds": 0.0009999999999994458,
          "url": "audio/podcast-05.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 696320,
          "duration": 15.789569160997733,
          "sha256": "b2e208e6ac1bff26df5b98f3d0396a2fdae36596ecf55422f0b488837a31d2da",
          "pcmSha256": "35ecd6ac7a99612055f010c8fd1111ea75f1097cebb0d35abe6a3f1d8b7c16c4",
          "peak": 0.9991886615753174,
          "rms": 0.14785876214067106,
          "durationDeltaSeconds": 0.030569160997732325,
          "url": "audio/podcast-05.fish.wav"
        },
        "original": {
          "sampleRate": 48000,
          "numSamples": 756432,
          "duration": 15.759,
          "sha256": "28e06bc8403524d75a475e2c05a3ce916d20d11b345a882a92c4369e5b5c19f6",
          "pcmSha256": "2ab64f1668212f4ef0312aa42991357d764a169010c559ef4ff707322654160a",
          "peak": 0.5968784093856812,
          "rms": 0.07713353985524649,
          "url": "audio/podcast-05.original.wav"
        }
      }
    },
    {
      "id": "h3-01",
      "corpus": "h3",
      "run": "h3-minimax-v1-20261008",
      "jobId": "2ed8aa278bab2615da6df244626f2a7ff52533a43fc1aa21d7d738645e15d4cb",
      "sampleId": "94229993d0ea1cdd87dbdeb49ce5440feac8566af87d9aee7d8bd9640cb266b9",
      "selectionRank": 290375,
      "inputVariant": "original",
      "source": {
        "uri": "r2://synthttsdata/minimaxh3_v1/001485_8147667e_take0_seed219794443.mp3",
        "objectSha256": "a63fcdf1063a2414323f349c2b76166cad8c3e3b0f3f3bf984446b8e895e4877",
        "startSample": 0,
        "sampleRate": 32000,
        "numSamples": 529920,
        "pcmSha256": "4f148ebfabe35ca8011a749ea02e19c8a22cc53b886f687194e6c8b8ed69055c",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/qwen/d590930fb07d4c7274ed7e01a1aeddfb.npz",
          "shard_sha256": "c284614ea9a5e268d41a945ed3fa7ebeb846fc3f55aea72a000ce31384b71343",
          "record_index": 35187,
          "sample_id": "94229993d0ea1cdd87dbdeb49ce5440feac8566af87d9aee7d8bd9640cb266b9",
          "token_shape": [
            16,
            207
          ],
          "tokens_npy_sha256": "8bd8898a788a8c31d293d54a33ebfcef0cb6671996806caa4a9c8cdbf7363541",
          "tokens_payload_sha256": "e88de96fb84aea988cf940a4c10a2533f6518996bf2539cc8eccf21281bf00d4",
          "downloadedAtUnix": 1791563020.4101071,
          "shardBytes": 490555565,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "qwen",
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 0,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/chatterbox/cb1d167d160f385a34cd073f94b49709.npz",
          "shard_sha256": "22c32582b2a9169c69a500bb9b8a0d9b33adaa10a4d2f28a28ab3a1de6be6bd4",
          "record_index": 166045,
          "sample_id": "94229993d0ea1cdd87dbdeb49ce5440feac8566af87d9aee7d8bd9640cb266b9",
          "token_shape": [
            1,
            414
          ],
          "tokens_npy_sha256": "bfc69e004b4f477445851908cd84d425bb98f26adcb558fad05f002c13d6057b",
          "tokens_payload_sha256": "750ea96fb485fa5b0e7d5d84bb641dfe14ca7f878e719da59e5401ffe7f77502",
          "downloadedAtUnix": 1791563084.124599,
          "shardBytes": 495808118,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "chatterbox",
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/fish/1ca17f09ccb391193284bc94c7f766b1.npz",
          "shard_sha256": "4b32a5af632913707f5e5ef510a094602431f2aac2bfcb7a6a0401659c018b45",
          "record_index": 16497,
          "sample_id": "94229993d0ea1cdd87dbdeb49ce5440feac8566af87d9aee7d8bd9640cb266b9",
          "token_shape": [
            10,
            357
          ],
          "tokens_npy_sha256": "b285c6717288035e88b8219ff51adc9fcff30b552148dd3678a2d769c490a9e6",
          "tokens_payload_sha256": "db586e154ee365284e1a6fe17b073a60fe75d45f916ff60d9f10138c7518c868",
          "downloadedAtUnix": 1791563030.648258,
          "shardBytes": 497903586,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fish",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 730296,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 397440,
          "duration": 16.56,
          "sha256": "7fa57c209eb8df553537e4e65667960edc4f2e72c3a9b3933cf4a019d730843b",
          "pcmSha256": "2e269dc611d0c8cca716707ea85bc35c567b1fde37a84cfd95ec794b1fc2107b",
          "peak": 0.6464089751243591,
          "rms": 0.029725157514177553,
          "durationDeltaSeconds": 0.0,
          "url": "audio/h3-01.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 397440,
          "duration": 16.56,
          "sha256": "c338486e1a31a9338dc588a668709944ea2ac3ffb0c9cf4adc44cd8d5e5e84a1",
          "pcmSha256": "c47cd5a8f44d9440948736b43b7bd1634c85e360ae28d7093416fc72607e2eb7",
          "peak": 0.9900000095367432,
          "rms": 0.054449359284097186,
          "durationDeltaSeconds": 0.0,
          "url": "audio/h3-01.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 731136,
          "duration": 16.579047619047618,
          "sha256": "6fc37292b1ce3b181851d3c0eba4b013d32af4f12aa34ebb153b4f948de34cc4",
          "pcmSha256": "8a37f0a65386d91fa695e9e5bc0f688c5b3f74f02367754cabd0d276b8e0c7a3",
          "peak": 0.9987204074859619,
          "rms": 0.08525102275165057,
          "durationDeltaSeconds": 0.01904761904761898,
          "url": "audio/h3-01.fish.wav"
        },
        "original": {
          "sampleRate": 32000,
          "numSamples": 529920,
          "duration": 16.56,
          "sha256": "47b40b3816e9f854af64a0cef30f8c515b93bba873f40ccd910b15180b3a64a2",
          "pcmSha256": "4f148ebfabe35ca8011a749ea02e19c8a22cc53b886f687194e6c8b8ed69055c",
          "peak": 0.8134092092514038,
          "rms": 0.04241278141258987,
          "url": "audio/h3-01.original.wav"
        }
      }
    },
    {
      "id": "h3-02",
      "corpus": "h3",
      "run": "h3-minimax-v1-20261008",
      "jobId": "093cd7bd9d03efe5ec8933df662423dec7885ea953de68850839e6949707a8ec",
      "sampleId": "a6affc8402964f5405db49605230fc1e2edac5b916a3551b6dd757a98f304052",
      "selectionRank": 56484,
      "inputVariant": "original",
      "source": {
        "uri": "r2://synthttsdata/minimaxh3_v1/000274_e6690c64_take1_seed9684656.mp3",
        "objectSha256": "27db74d288fdd6b175328dd069a2263f43191f7c7282cb1add7caa04cea1c884",
        "startSample": 502848,
        "sampleRate": 32000,
        "numSamples": 502848,
        "pcmSha256": "d9e260490d87f4903cbcfe3d002f706ef16e43ff34faaed04b2b1906634e14f3",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/qwen/53b047bf6903b7c6c5c774dcb1e879a9.npz",
          "shard_sha256": "65b63a4d76df90ead6ad874c4973cb19952f6c562a7835f16a13d23b688f21d8",
          "record_index": 2046,
          "sample_id": "a6affc8402964f5405db49605230fc1e2edac5b916a3551b6dd757a98f304052",
          "token_shape": [
            16,
            197
          ],
          "tokens_npy_sha256": "06d0956d650e6b515064ccce1a762afdaebab9157635976bb87512dcc283eb32",
          "tokens_payload_sha256": "c8f88b8cf2fdb41cd1fe7b47da9940f595863e66ed8b526ba5d06721b255b0a0",
          "downloadedAtUnix": 1791563063.067545,
          "shardBytes": 490139988,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "qwen",
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1104,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/chatterbox/087acfffb9be53412bc70a9cb1d2476e.npz",
          "shard_sha256": "23492651994abe60a01dff1e1d2f4d6a0920178caad64ea84cc65080dc4b52f8",
          "record_index": 67568,
          "sample_id": "a6affc8402964f5405db49605230fc1e2edac5b916a3551b6dd757a98f304052",
          "token_shape": [
            1,
            393
          ],
          "tokens_npy_sha256": "edf804cbe2be236f0286889e0bc4298c003e56569db2661ea61b00dd64431b06",
          "tokens_payload_sha256": "4a04f71855412ce8c7af6264b94901f6ae36c2bda769ec430cdf882e34df49bf",
          "downloadedAtUnix": 1791563104.8238108,
          "shardBytes": 492701863,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "chatterbox",
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/fish/08a6376db6ee1357a70365be6a1dbcff.npz",
          "shard_sha256": "c8e5159b5ba0ad2133a0d6df0274d3db09e290b813b4b48cc9cc35a981718f6b",
          "record_index": 15336,
          "sample_id": "a6affc8402964f5405db49605230fc1e2edac5b916a3551b6dd757a98f304052",
          "token_shape": [
            10,
            339
          ],
          "tokens_npy_sha256": "6ecf1225bb314960244a108333dcea946e2ab6629fbd6c52c964dfa520ffb6aa",
          "tokens_payload_sha256": "38de406e509d27fe671d34e1f4f51ef9bfb7edfdbf9125b704dcb969251f847e",
          "downloadedAtUnix": 1791563088.103001,
          "shardBytes": 498105519,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fish",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 692987,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 378240,
          "duration": 15.76,
          "sha256": "d95843ab5b5320ada8cfa9327865ae4b9f535a4924816d6ff449481e3ad19a96",
          "pcmSha256": "1129f814301cb9f4bd31e527ba5690ba9d79aa5d0c521ae83456cdead28b807b",
          "peak": 0.7026164531707764,
          "rms": 0.07296348283132771,
          "durationDeltaSeconds": 0.045999999999999375,
          "url": "audio/h3-02.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 377280,
          "duration": 15.72,
          "sha256": "6296f231cbbde97bb73c13244395baf6f4cf10cd578e89614292c70c5187f9b0",
          "pcmSha256": "3c907a4fd2695fcd950b8dfc3430c3d009f96ff621b07edf5b0da26e0479c4b6",
          "peak": 0.7521596550941467,
          "rms": 0.08112705025659242,
          "durationDeltaSeconds": 0.006000000000000227,
          "url": "audio/h3-02.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 694272,
          "duration": 15.74312925170068,
          "sha256": "fa77d33f8b4d74fcf590bdcb3fba4a8463e83665d94c0030b2bd1201e29789d9",
          "pcmSha256": "efdef3c5061848c15b86ccf5e7dae80fbd17240eb1fc66c66d64884b0a85080e",
          "peak": 0.9992548823356628,
          "rms": 0.12775789064871745,
          "durationDeltaSeconds": 0.02912925170067915,
          "url": "audio/h3-02.fish.wav"
        },
        "original": {
          "sampleRate": 32000,
          "numSamples": 502848,
          "duration": 15.714,
          "sha256": "15e76ae3ee0931a1d193b85217737d9ca55fe2d13bba7811cf52c8236adc8d2d",
          "pcmSha256": "d9e260490d87f4903cbcfe3d002f706ef16e43ff34faaed04b2b1906634e14f3",
          "peak": 0.6610528230667114,
          "rms": 0.0771431649176511,
          "url": "audio/h3-02.original.wav"
        }
      }
    },
    {
      "id": "h3-03",
      "corpus": "h3",
      "run": "h3-minimax-v1-20261008",
      "jobId": "24ad37d6181ba9546e5f509f389c29058375af01c97e5c4cae4cb2d46cc680d6",
      "sampleId": "72c6f503003148bfe303e2a64434fc79a681cc552b77160a10d04c6bb62d094e",
      "selectionRank": 228728,
      "inputVariant": "original",
      "source": {
        "uri": "r2://synthttsdata/minimaxh3_v1/002976_f79ce909_take0_seed2145746504.mp3",
        "objectSha256": "0d6e6c227fdee928de6c41b9b1400267ee5168a7b1c2a7241e6a7b256cd98445",
        "startSample": 0,
        "sampleRate": 32000,
        "numSamples": 574848,
        "pcmSha256": "1fb7482d1b1c2b8693cf9060fed894d7c39050f490848be850e2f37bc2427a27",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/qwen/d909af4fbcc4ba8aa69b07ecac25440a.npz",
          "shard_sha256": "daf4c0dc06f1e2a7971f2242d98762d771069df78ce6821161ee086476346504",
          "record_index": 11290,
          "sample_id": "72c6f503003148bfe303e2a64434fc79a681cc552b77160a10d04c6bb62d094e",
          "token_shape": [
            16,
            225
          ],
          "tokens_npy_sha256": "35bbc648c776c20075cf575a484e8d7229ffbe00adcefcde0cd790f179f399bc",
          "tokens_payload_sha256": "c9e93783a7317cb3d62d39e4f010e4817b9d6a5f2439cc3d608e09ddbbed93a4",
          "downloadedAtUnix": 1791563072.1140814,
          "shardBytes": 160965938,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "qwen",
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 864,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/chatterbox/9f40b389d261a0466ee3811b215e84ba.npz",
          "shard_sha256": "87171f2c9cd49dc7e2e36be983d017c8d7b84fe44821d1de4bafcc8f85930170",
          "record_index": 65675,
          "sample_id": "72c6f503003148bfe303e2a64434fc79a681cc552b77160a10d04c6bb62d094e",
          "token_shape": [
            1,
            449
          ],
          "tokens_npy_sha256": "798c7961ed04d4859f5581f3606aab921c37844595eea3443dc26481f09168c9",
          "tokens_payload_sha256": "9e12f4cf2bdde6869abce6e22e423c41bf54666beb631dee566f4a8e00155e34",
          "downloadedAtUnix": 1791563077.0800092,
          "shardBytes": 129917660,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "chatterbox",
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/fish/93248c0f756cf07b309e997b621a8f1d.npz",
          "shard_sha256": "31efdf6c0f69864d50eaf952b78e009c057f09e70e20c9f7c09a8ff956daa246",
          "record_index": 855,
          "sample_id": "72c6f503003148bfe303e2a64434fc79a681cc552b77160a10d04c6bb62d094e",
          "token_shape": [
            10,
            387
          ],
          "tokens_npy_sha256": "149dc1a773275f4a689c030963747f9d950e1385b89c3ab88afc8477e8b6d82b",
          "tokens_payload_sha256": "c4582d6a1897247c8f52da04462f892dd577d3225b7f5716a841a5a2988cef98",
          "downloadedAtUnix": 1791563068.9573817,
          "shardBytes": 88278690,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fish",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 792212,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 432000,
          "duration": 18.0,
          "sha256": "ae83753279f05b1226283c8e22064d1b112e43259554c6b2cb56a48467435f52",
          "pcmSha256": "866031f97e23c1fe4538473efdc24fd3d99afe9c1c604439c3722200887bfee6",
          "peak": 0.6486212611198425,
          "rms": 0.020723889186806548,
          "durationDeltaSeconds": 0.036000000000001364,
          "url": "audio/h3-03.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 431040,
          "duration": 17.96,
          "sha256": "38f2096d3c48df6c1c1a956af60bd0c1af400b9d57a811494683073672a157ad",
          "pcmSha256": "1ccc36d3dbab4f659a566cc295d1aa6def6c54227e87542326cb8a0b3f617612",
          "peak": 0.9261171817779541,
          "rms": 0.03211598634716864,
          "durationDeltaSeconds": -0.003999999999997783,
          "url": "audio/h3-03.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 792576,
          "duration": 17.972244897959182,
          "sha256": "5b19623449251931178201ae9b670aca7d64b5809e41be34f3349ad67deb6363",
          "pcmSha256": "6dddde374b82e42d165d67cceb6fd00be0b79b06ded0349cf44126784231756e",
          "peak": 0.9992256164550781,
          "rms": 0.05142655185913687,
          "durationDeltaSeconds": 0.008244897959183817,
          "url": "audio/h3-03.fish.wav"
        },
        "original": {
          "sampleRate": 32000,
          "numSamples": 574848,
          "duration": 17.964,
          "sha256": "6ac4f4776850540e50419cefef4324850fff1a97f462f9a14be9f8074d3c0bb5",
          "pcmSha256": "1fb7482d1b1c2b8693cf9060fed894d7c39050f490848be850e2f37bc2427a27",
          "peak": 0.8897804021835327,
          "rms": 0.030253742790167144,
          "url": "audio/h3-03.original.wav"
        }
      }
    },
    {
      "id": "h3-04",
      "corpus": "h3",
      "run": "h3-minimax-v1-20261008",
      "jobId": "18660c0c0e3fed7685d84f854bf55bc6aeaf8fff8f09291a0abfe50de92016e0",
      "sampleId": "f319750ea82e3be53f9f1108175ca8ff0ec2b31a0a6db44dc8dd04263f235d9b",
      "selectionRank": 156840,
      "inputVariant": "original",
      "source": {
        "uri": "r2://synthttsdata/minimaxh3_v1/000669_eb7ec41c_take1_seed1238904690.mp3",
        "objectSha256": "303074cb4e80671c11f02c91ddf44b8643c220c7e0dc4f6ce36dd1ed919274d7",
        "startSample": 502848,
        "sampleRate": 32000,
        "numSamples": 502848,
        "pcmSha256": "e3e8cfa5b1da3e4fcd349bf020ec8fd6444a9060afc059722214fc1f75a36bb6",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/qwen/de4496436a0f63b26d7828ae3dcd1560.npz",
          "shard_sha256": "7caed12a5f538d36446772c75515d18aeffa160dd46d74aa798a6ed21b58b098",
          "record_index": 51502,
          "sample_id": "f319750ea82e3be53f9f1108175ca8ff0ec2b31a0a6db44dc8dd04263f235d9b",
          "token_shape": [
            16,
            197
          ],
          "tokens_npy_sha256": "58245492ffea43730975ec15b0bf9c25048e6ea2fb82db9189762b1f892701bd",
          "tokens_payload_sha256": "5e52dc2be204d804fd3b06bffe553be30cd6ca6d88a0156dd3efb7bc7b29f001",
          "downloadedAtUnix": 1791563123.5786448,
          "shardBytes": 493716850,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "qwen",
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 1104,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/chatterbox/bd9c28479b78a1f6ae6773be152f8bfe.npz",
          "shard_sha256": "3c824e250730fbef0204bc57651644f540de21d08ea9794039d002dd059d448e",
          "record_index": 186822,
          "sample_id": "f319750ea82e3be53f9f1108175ca8ff0ec2b31a0a6db44dc8dd04263f235d9b",
          "token_shape": [
            1,
            393
          ],
          "tokens_npy_sha256": "d3cdc199c668733e62e7064db070adc1fe08f1bbe14a2feb59f5536dc1a8fe2b",
          "tokens_payload_sha256": "a1323c38367b74cc27371e38aeeaac4cc52749fbc451b6e9685d96fdf0e56659",
          "downloadedAtUnix": 1791563125.3368306,
          "shardBytes": 494137119,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "chatterbox",
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/fish/e4270e40c913f90f99db3dc26f29f0e5.npz",
          "shard_sha256": "3c735704bfdad03a19538685d5145d904dde6335897591a2ffd1fdaa5cadf6f5",
          "record_index": 21828,
          "sample_id": "f319750ea82e3be53f9f1108175ca8ff0ec2b31a0a6db44dc8dd04263f235d9b",
          "token_shape": [
            10,
            339
          ],
          "tokens_npy_sha256": "4face2b4be2780b99bad1318fac37e4c3a0eadd1966b397bb952eb76e833f0a8",
          "tokens_payload_sha256": "d99fdfaefe5bfb0b8ee4d7e7e4362999bffc75db1f6b3d112eb3e2533a803ba4",
          "downloadedAtUnix": 1791563122.8624167,
          "shardBytes": 498229697,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fish",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 692987,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 378240,
          "duration": 15.76,
          "sha256": "3e01477ad004081bd90044e881e2d645b95abeab55c4a90633230935f758f6f6",
          "pcmSha256": "c791a895e2a491fa90fae0c79fbf1e99cc12c73d5176dca058bd64e7bb72341a",
          "peak": 0.7971764802932739,
          "rms": 0.10798530308176643,
          "durationDeltaSeconds": 0.045999999999999375,
          "url": "audio/h3-04.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 377280,
          "duration": 15.72,
          "sha256": "0119ad7763414a3c61c3660e72fb08f9f8e297763dae77d6bfb719cc8da04419",
          "pcmSha256": "d2979d4e8a8e894b45a6b1b25df89cd7a40e247e2f7cf32c89a2289840489912",
          "peak": 0.9495334029197693,
          "rms": 0.11869383961597173,
          "durationDeltaSeconds": 0.006000000000000227,
          "url": "audio/h3-04.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 694272,
          "duration": 15.74312925170068,
          "sha256": "b0fd6a9d42ad71f64449883ed55ddb35b3687ae174b0942646b8f70324e565f7",
          "pcmSha256": "2b2374c9a7e9c348575d7e15bda56fa1e56c856aa35653c0951b983965e0f6e1",
          "peak": 0.9987484812736511,
          "rms": 0.13707835435239635,
          "durationDeltaSeconds": 0.02912925170067915,
          "url": "audio/h3-04.fish.wav"
        },
        "original": {
          "sampleRate": 32000,
          "numSamples": 502848,
          "duration": 15.714,
          "sha256": "bf83ed0d0df4d0c0c1959c1e56944db0539d910d1da8b0f55bd66fec58127a4d",
          "pcmSha256": "e3e8cfa5b1da3e4fcd349bf020ec8fd6444a9060afc059722214fc1f75a36bb6",
          "peak": 0.7977741956710815,
          "rms": 0.11264797429849462,
          "url": "audio/h3-04.original.wav"
        }
      }
    },
    {
      "id": "h3-05",
      "corpus": "h3",
      "run": "h3-minimax-v1-20261008",
      "jobId": "089d7e30630fe38eb250738e9699d787affebad0d372b96ea02c85d853b85e34",
      "sampleId": "67c5d1fb59e1539dac8c353a726a3408d6dae62ae196c1a115123e3e5a48dbde",
      "selectionRank": 52973,
      "inputVariant": "original",
      "source": {
        "uri": "r2://synthttsdata/minimaxh3_v1/002206_040d517b_take0_seed1436398831.mp3",
        "objectSha256": "a39e29879612e72045350b6dbe9b4ba8af5770cd867f6cdaca3bdc7033cfa693",
        "startSample": 0,
        "sampleRate": 32000,
        "numSamples": 347904,
        "pcmSha256": "fdd3466e8f21bc724b84f58de3c26d25d46d959241948ab104c74e9ba53e8ec7",
        "exactEncoderInputVerified": true
      },
      "tokens": {
        "qwen": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/qwen/92982a60039b613546ffc2aa5ad0a607.npz",
          "shard_sha256": "479dca2ce434ef2d13e762073e9fdb1acb14b73e47e9425867d93b82da07112c",
          "record_index": 2224,
          "sample_id": "67c5d1fb59e1539dac8c353a726a3408d6dae62ae196c1a115123e3e5a48dbde",
          "token_shape": [
            16,
            136
          ],
          "tokens_npy_sha256": "cc449ba891596fca4f2a16c8243e7840a24bb08f7715eee31fa09b73410c8d22",
          "tokens_payload_sha256": "62f81e346e4a5dee065a3062f9220aba710ccfe3ca910d5c4f27973a33542db6",
          "downloadedAtUnix": 1791563133.3758519,
          "shardBytes": 499324678,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "qwen",
            "model_id": "Qwen/Qwen3-TTS-Tokenizer-12Hz",
            "revision": "7dd38ad4e9bad454aae9cd937d0cd577604fe229",
            "checkpoint_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258",
            "source_commit": "022e286b98fbec7e1e916cb940cdf532cd9f488e",
            "num_codebooks": 16,
            "internal_sample_rate": 24000,
            "normalization": "none; no amplitude scaling or clipping",
            "precision": "float32; CUDA matmul and cuDNN TF32 disabled",
            "resampling": "librosa.resample(res_type='soxr_hq', fix=True, scale=False)"
          },
          "decoding": {
            "all_saved_codebooks_used": 16,
            "token_values_unchanged": true,
            "applied_alignment_shift_samples": 0,
            "applied_trim_samples": 0,
            "frame_tail_samples_24k": 192,
            "samples_at_native_clamp": 0
          }
        },
        "chatterbox": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/chatterbox/087acfffb9be53412bc70a9cb1d2476e.npz",
          "shard_sha256": "23492651994abe60a01dff1e1d2f4d6a0920178caad64ea84cc65080dc4b52f8",
          "record_index": 288392,
          "sample_id": "67c5d1fb59e1539dac8c353a726a3408d6dae62ae196c1a115123e3e5a48dbde",
          "token_shape": [
            1,
            272
          ],
          "tokens_npy_sha256": "959f2a9b0cd0a680ebefa11477637f071c60c3689148ee1d42647f41d5de5600",
          "tokens_payload_sha256": "7020def64978732c3d86579eb197abd019cfb0702f56ccb50ea1b23c7d197514",
          "downloadedAtUnix": 1791563104.8238108,
          "shardBytes": 492701863,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "chatterbox",
            "checkpoint_sha256": "2b78103c654207393955e4900aac14a12de8ef25f4b09424f1ef91941f161d4e",
            "checkpoint_revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
            "tokenizer_sha256": "1641b71a0059b74f37f0c56e4200f3d8b28fcff0fd0328ba452ebfc776db4564",
            "implementation_revision": "5de7a54aa4e5e2baadb0182dde554908b48b85c2",
            "num_codebooks": 1,
            "frame_rate": 25,
            "normalization": {
              "log_mel": "per clip: log10(max(mel_power,1e-10)); clamp at clip maximum - 8; (x+4)/4",
              "mel_filters": "128-bin filters loaded from checkpoint tokenizer._mel_filters",
              "waveform": "none"
            },
            "resampling": "librosa.resample(res_type=soxr_hq, fix=True, scale=False)"
          },
          "decoding": {
            "conditioning": {
              "source": "same original clip, beginning only",
              "seconds": 10.0,
              "num_samples_24k": 240000,
              "stored_prompt_tokens": 250,
              "prompt_is_exact_saved_content_prefix": true,
              "newly_encoded_prompt_tokens": 0,
              "speaker_embedding": "original S3Gen CAMPPlus from reference resampled to 16 kHz",
              "reference_mels": "original S3Gen 24 kHz mel extractor; 2 mel frames per saved prompt token",
              "resampling": "librosa soxr_hq original→24k; torchaudio default sinc 24k→16k",
              "limitation": "Reference-conditioned synthesis from stored tokens, not a self-contained waveform codec reconstruction."
            },
            "tokenizer_forward_calls": 0,
            "seed": 20261009,
            "cfm_timesteps": 10,
            "official_decoder_output_fade": "upstream 20 ms zero prefix + 20 ms cosine fade retained",
            "alignment_or_loudness_processing_applied": false
          }
        },
        "fish": {
          "r2_bucket": "special-tok",
          "r2_key": "h3-minimax-v1-20261008/fish/fce302d6006eadd37f11ef1c48ce3926.npz",
          "shard_sha256": "f268893c0c1b6de24e9332b8fe02c09b58427fd4718d65f641647f211b05cc9c",
          "record_index": 23643,
          "sample_id": "67c5d1fb59e1539dac8c353a726a3408d6dae62ae196c1a115123e3e5a48dbde",
          "token_shape": [
            10,
            235
          ],
          "tokens_npy_sha256": "a56ab7fdf3a5a4f2f1fbeeae72c3d9d98b3dec903c531de01954eb5d5568f23c",
          "tokens_payload_sha256": "ff677264d0a35376d77f9a52c40f7e14b89b89515dcea562965b6a4db74e5b7d",
          "downloadedAtUnix": 1791563141.880215,
          "shardBytes": 490543552,
          "downloadVerification": "fresh_r2_get_full_file_sha256",
          "encodingMetadata": {
            "codec": "fish",
            "revision": "step5600000-export20261006",
            "checkpoint": "s3://artifacts/fishdac_step5600000_inference_20261006.tar.zst#checkpoint/generator.pth",
            "checkpoint_sha256": "6e761a33ebce9e99a59b06f912fcde02ab1063c0dcd939964b537905d93f24db",
            "source_sha256": {
              "code/dac/__init__.py": "f78293bdb74e1b05db56266e3a42e07ff52baef9e26bbac89a5849e667eca867",
              "code/dac/model/__init__.py": "4b6d119cca43d2dc1c80122d8ce5a5f88f1a27ac9db38945c6dde45b31956435",
              "code/dac/model/base.py": "2361726838faae352f1265d514e3e2f8d978a2b278dcf66bb5799dfbd2ba7b83",
              "code/dac/model/dac.py": "53951a9defbb32092f02ba0ed206eb424c6a9a1536dfb85abd9c9583d066a343",
              "code/dac/nn/__init__.py": "23ba6d87b964da0f35eb9ce8c8917c57e1738520172048cfd75a331e3a9a892b",
              "code/dac/nn/bottleneck.py": "3d193a533b32cd466f70ee263317fd5d18f2031f70892753fb5958ea5ba6c444",
              "code/dac/nn/fsq.py": "6860060f133a44c42b99320a5952d0f1b891acc176534c1cb1b26f60e7bc61b8",
              "code/dac/nn/layers.py": "ec2649649d787b166a138d5ad9dcd585aeabbcd93670771b30cbbbab731c4b63",
              "code/dac/nn/quantize.py": "2a0d7ce5e258560c28926bdf63d5878e5fb08156497fa4d9b15906114a4f7585",
              "code/dac/nn/semantic.py": "57822fedfd1f5dc2c79dfe4c24b391d9b209988e2f1474f8041d5cf4e93d27af",
              "code/dac/utils/__init__.py": "ac4e097f5db5f730bdd944bfd7ddf66761e776ba22d14d5ad7e3d6bd902addf5"
            },
            "codebook_sizes": [
              4096,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024,
              1024
            ],
            "num_codebooks": 10,
            "frame_rate": 21.533203125,
            "hop_length": 2048,
            "internal_sample_rate": 44100,
            "normalization": "per-clip CPU AudioSignal.normalize(-16.0) LUFS",
            "precision": "float32; cuDNN TF32 enabled (bundled reference default); matmul TF32 disabled; no autocast",
            "resampling": "julius.ResampleFrac, exact AudioSignal.resample implementation; floor(N * 44100 / source_rate)"
          },
          "decoding": {
            "codebooks_decoded": 10,
            "expected_native_samples": 479455,
            "reference_verification": null
          }
        }
      },
      "outputs": {
        "qwen": {
          "sampleRate": 24000,
          "numSamples": 261120,
          "duration": 10.88,
          "sha256": "fe99221e3b785076e5ecb4a80ab139fa5c8156726a466a22f6e6ceeeabb2fd51",
          "pcmSha256": "a3604421920dcab1dc8af832b835d5ebaaa986f7991daf473baef4a8b0b5c2ba",
          "peak": 0.3951444625854492,
          "rms": 0.03758009086419668,
          "durationDeltaSeconds": 0.008000000000000895,
          "url": "audio/h3-05.qwen.wav"
        },
        "chatterbox": {
          "sampleRate": 24000,
          "numSamples": 261120,
          "duration": 10.88,
          "sha256": "9c729b9e87bc2d86921533eb9e3f1b18703b662f80f4a7ed027a4a55686c8016",
          "pcmSha256": "ac0d2866c04816478151494c763b58ea3c445f24e43ef0be1bfc7710b450c7e6",
          "peak": 0.3618147075176239,
          "rms": 0.0349820462704853,
          "durationDeltaSeconds": 0.008000000000000895,
          "url": "audio/h3-05.chatterbox.wav"
        },
        "fish": {
          "sampleRate": 44100,
          "numSamples": 481280,
          "duration": 10.913378684807256,
          "sha256": "d87b5b644702199a1db280f8c7cf7bdaee297931e1bce19541ba677cb494899e",
          "pcmSha256": "583fa3c40798547995b689e45c70549fdb2caee7bdafa927ebbf46ac584b181c",
          "peak": 0.99898761510849,
          "rms": 0.11636419826012653,
          "durationDeltaSeconds": 0.04137868480725615,
          "url": "audio/h3-05.fish.wav"
        },
        "original": {
          "sampleRate": 32000,
          "numSamples": 347904,
          "duration": 10.872,
          "sha256": "89e6533edc8cbc4e665a52124f11a03f89547596c2de6986de97e6b82cc4d6d7",
          "pcmSha256": "fdd3466e8f21bc724b84f58de3c26d25d46d959241948ab104c74e9ba53e8ec7",
          "peak": 0.38809633255004883,
          "rms": 0.03920393968582107,
          "url": "audio/h3-05.original.wav"
        }
      }
    }
  ]
}