{
  "created_at": "2026-09-15T14:50:59+03:00",
  "status": "completed",
  "instance_id": 50735875,
  "gpu": "NVIDIA A100-SXM4-40GB",
  "dataset": "Seed-TTS-Eval English",
  "samples_per_model": 1088,
  "models": 38,
  "scored_audio": 41287,
  "completed_shards": 198,
  "dataset_sha256": "8a9386efb1768a90ffba66f8931e3887b53e9c9d1bf246eb46a8cce32bc7b1c1",
  "suite_sha256": "1a37627704da3f0158d29a0cc3cff712494dadd0f03553fea2511a01d60ea8a9",
  "source_commit": "3ce0a0aa9cf3c9b24d8e4d6c3d0315b7bf7adc0a",
  "asr": {
    "checkpoint": "Systran/faster-whisper-large-v3",
    "revision": "edaa852ec7e145841d8ffdb056a99866b5f0a478",
    "device": "cuda",
    "compute_type": "float16",
    "language": "en",
    "beam_size": 5,
    "temperature": 0,
    "condition_on_previous_text": false,
    "vad_filter": false
  },
  "normalization": "whisper_english",
  "seed": 42,
  "repeats": 1,
  "ci_method": "95% prompt-cluster bootstrap; 1000 resamples; RNG seed 42; full 1088 prompts",
  "protocol": "Fixed provider voice/reference intelligibility; not official zero-shot speaker identity/SIM evaluation",
  "backup_verified_at": "2026-09-15T11:48:02.499471+00:00",
  "dataset_id": "kadirnar/voicehub-arena-seed-tts-eval",
  "dataset_revision": "56c87034f5ef6a4b7bcfe403390ed86a0fc36611",
  "table": [
    {
      "model": "kokoro",
      "name": "Kokoro",
      "checkpoint": "hexgrad/Kokoro-82M",
      "revision": "cbc78411372edb46f7e42030b241834a55fe0cb6",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.007682849762477258,
        0.011640828342247611
      ],
      "cer_ci95": [
        0.001896678239861543,
        0.0029770741117294947
      ],
      "wer": 0.009629071422590639,
      "cer": 0.0024400404689638756,
      "mer": 0.00962423633776885,
      "wil": 0.01742962890275368,
      "wip": 0.9825703710972463,
      "exact_match_rate": 0.9136029411764706,
      "utterance_mean_wer": 0.009572094061983769,
      "utterance_mean_cer": 0.002414540728261279,
      "rtf": 0.018666759712283744,
      "latency_p50_s": 0.08802692033350468,
      "latency_p95_s": 0.10426744297146796,
      "peak_vram_mib": 478.27685546875,
      "audio_seconds": 4623.275,
      "silence_ratio": 0.3701472327051314,
      "clipping_ratio": 0.0,
      "rms_dbfs": -26.425495242960476,
      "protocol": "Original provider configuration"
    },
    {
      "model": "supertonic",
      "name": "Supertonic",
      "checkpoint": "Supertone/supertonic-3",
      "revision": "3cadd1ee6394adea1bd021217a0e650ede09a323",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.00785054517897158,
        0.011805873826184896
      ],
      "cer_ci95": [
        0.002118013958236292,
        0.003225100490647234
      ],
      "wer": 0.009796533534287867,
      "cer": 0.002663214902100815,
      "mer": 0.009789156626506024,
      "wil": 0.017839658400852665,
      "wip": 0.9821603415991473,
      "exact_match_rate": 0.9099264705882353,
      "utterance_mean_wer": 0.009449587931955387,
      "utterance_mean_cer": 0.002529207763588537,
      "rtf": 0.06423751407986668,
      "latency_p50_s": 0.2797527089715004,
      "latency_p95_s": 0.3004460956901312,
      "peak_vram_mib": 419.396484375,
      "audio_seconds": 4756.657505668934,
      "silence_ratio": 0.43568262751000425,
      "clipping_ratio": 0.0,
      "rms_dbfs": -26.68671776510351,
      "protocol": "Original provider configuration"
    },
    {
      "model": "omnivoice",
      "name": "OmniVoice",
      "checkpoint": "k2-fsa/OmniVoice",
      "revision": "c5fdb5ccb189668d56333f77ba2629f4cd7535f4",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.007861317788115801,
        0.012080484304594895
      ],
      "cer_ci95": [
        0.0018904921536443545,
        0.003657155632574282
      ],
      "wer": 0.009880264590136481,
      "cer": 0.002648336606558353,
      "mer": 0.009874476987447698,
      "wil": 0.0173500167374373,
      "wip": 0.9826499832625627,
      "exact_match_rate": 0.9154411764705882,
      "utterance_mean_wer": 0.009585752278766985,
      "utterance_mean_cer": 0.0024681703906616025,
      "rtf": 0.328981732833577,
      "latency_p50_s": 1.332394178956747,
      "latency_p95_s": 1.5358351096510887,
      "peak_vram_mib": 3228.8759765625,
      "audio_seconds": 4436.8,
      "silence_ratio": 0.3348716952743415,
      "clipping_ratio": 0.0,
      "rms_dbfs": -23.852963743668997,
      "protocol": "Original provider configuration"
    },
    {
      "model": "speecht5",
      "name": "SpeechT5",
      "checkpoint": "microsoft/speecht5_tts",
      "revision": "30fcde30f19b87502b8435427b5f5068e401d5f6",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.008283238600798748,
        0.012222576854350246
      ],
      "cer_ci95": [
        0.00235856560816167,
        0.003572732489586244
      ],
      "wer": 0.010131457757682325,
      "cer": 0.0029459025174076057,
      "mer": 0.010123828647925033,
      "wil": 0.01858563704868821,
      "wip": 0.9814143629513118,
      "exact_match_rate": 0.90625,
      "utterance_mean_wer": 0.009766929608381814,
      "utterance_mean_cer": 0.002814133144678633,
      "rtf": 0.1622327560700419,
      "latency_p50_s": 0.6036052946001291,
      "latency_p95_s": 0.9177474867552518,
      "peak_vram_mib": 726.37158203125,
      "audio_seconds": 4101.024,
      "silence_ratio": 0.27561073823840104,
      "clipping_ratio": 0.0,
      "rms_dbfs": -25.413263645785655,
      "protocol": "Original provider configuration"
    },
    {
      "model": "f5tts",
      "name": "F5TTS",
      "checkpoint": "F5TTS_v1_Base",
      "revision": null,
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.008544055321581331,
        0.012617792351167754
      ],
      "cer_ci95": [
        0.002511205039574283,
        0.0039833394587379875
      ],
      "wer": 0.010550113036925395,
      "cer": 0.0031988335416294708,
      "mer": 0.010543933054393306,
      "wil": 0.018842870213572005,
      "wip": 0.981157129786428,
      "exact_match_rate": 0.9090073529411765,
      "utterance_mean_wer": 0.011274173130790778,
      "utterance_mean_cer": 0.004422374102850214,
      "rtf": 0.7815442867831554,
      "latency_p50_s": 2.7469177208840847,
      "latency_p95_s": 2.9705387711524964,
      "peak_vram_mib": 1516.4052734375,
      "audio_seconds": 3739.3813333333333,
      "silence_ratio": 0.21193145480339617,
      "clipping_ratio": 3.4988988015009135e-06,
      "rms_dbfs": -19.899939920591734,
      "protocol": "Original provider configuration"
    },
    {
      "model": "styletts2",
      "name": "StyleTTS2",
      "checkpoint": "artifacts/styletts2/epochs_2nd_00020.pth",
      "revision": null,
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.008889246341617306,
        0.01317792950459316
      ],
      "cer_ci95": [
        0.0023622418836487065,
        0.003664567204361421
      ],
      "wer": 0.010885037260319853,
      "cer": 0.0029905374040349935,
      "mer": 0.010876840696117804,
      "wil": 0.019831987068541723,
      "wip": 0.9801680129314583,
      "exact_match_rate": 0.9044117647058824,
      "utterance_mean_wer": 0.010743798174864353,
      "utterance_mean_cer": 0.0029116890397277357,
      "rtf": 0.04208604152148397,
      "latency_p50_s": 0.19795850105583668,
      "latency_p95_s": 0.2473351741209626,
      "peak_vram_mib": 1068.0185546875,
      "audio_seconds": 5058.533333333333,
      "silence_ratio": 0.29950891529922624,
      "clipping_ratio": 0.0,
      "rms_dbfs": -22.315291148181117,
      "protocol": "Original provider configuration"
    },
    {
      "model": "echo",
      "name": "EchoTTS",
      "checkpoint": "jordand/echo-tts-base",
      "revision": "06731f67000241d55cff6b0a961f22f5f151f24f",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.008904209796796562,
        0.013934323157998304
      ],
      "cer_ci95": [
        0.0025651236034547214,
        0.0047400288829906
      ],
      "wer": 0.01121996148371431,
      "cer": 0.0034815211569362613,
      "mer": 0.011207761793241887,
      "wil": 0.020324327635941386,
      "wip": 0.9796756723640586,
      "exact_match_rate": 0.9034926470588235,
      "utterance_mean_wer": 0.011593386444029826,
      "utterance_mean_cer": 0.003852356493516786,
      "rtf": 0.5723286716880129,
      "latency_p50_s": 2.456841016188264,
      "latency_p95_s": 2.5182850835844874,
      "peak_vram_mib": 9363.14453125,
      "audio_seconds": 4676.359546485261,
      "silence_ratio": 0.12887222618488867,
      "clipping_ratio": 0.0,
      "rms_dbfs": -13.952829189004829,
      "protocol": "Original provider configuration"
    },
    {
      "model": "fishtts",
      "name": "FishTTS",
      "checkpoint": "fishaudio/s2-pro",
      "revision": "1de9996b6be38b745688de084d87a5633f714e4e",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.009372852661418718,
        0.013923650739157823
      ],
      "cer_ci95": [
        0.0025164327337067866,
        0.003856638680410998
      ],
      "wer": 0.011471154651260152,
      "cer": 0.0031690769505445457,
      "mer": 0.011463475859760689,
      "wil": 0.020912083738239273,
      "wip": 0.9790879162617607,
      "exact_match_rate": 0.8970588235294118,
      "utterance_mean_wer": 0.010823033686085157,
      "utterance_mean_cer": 0.002976193604265695,
      "rtf": 3.7434938590228506,
      "latency_p50_s": 14.975895503535867,
      "latency_p95_s": 21.895659835077815,
      "peak_vram_mib": 14652.67431640625,
      "audio_seconds": 4357.735328798186,
      "silence_ratio": 0.13489419262981148,
      "clipping_ratio": 0.0,
      "rms_dbfs": -16.634614222894925,
      "protocol": "Original provider configuration"
    },
    {
      "model": "xtts",
      "name": "XTTS",
      "checkpoint": "artifacts/xtts2",
      "revision": null,
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.009479942028329125,
        0.013929717071918904
      ],
      "cer_ci95": [
        0.003047702271545646,
        0.004894855067600172
      ],
      "wer": 0.011722347818805995,
      "cer": 0.003927870023210141,
      "mer": 0.011700793982448809,
      "wil": 0.020155170204901096,
      "wip": 0.9798448297950989,
      "exact_match_rate": 0.8979779411764706,
      "utterance_mean_wer": 0.012030761823960354,
      "utterance_mean_cer": 0.003932957956043187,
      "rtf": 0.37659089175029953,
      "latency_p50_s": 1.377348417416215,
      "latency_p95_s": 2.1737355709075925,
      "peak_vram_mib": 2044.58837890625,
      "audio_seconds": 4109.9946666666665,
      "silence_ratio": 0.12863922070376485,
      "clipping_ratio": 0.0,
      "rms_dbfs": -17.494293341348765,
      "protocol": "Original provider configuration"
    },
    {
      "model": "zonos2",
      "name": "Zonos2",
      "checkpoint": "Zyphra/ZONOS2",
      "revision": "65f1e80f94b599d474bb6af9094a803dc52f60bd",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.009429594420391598,
        0.01437212173948811
      ],
      "cer_ci95": [
        0.0030648629407874664,
        0.0057465328607648365
      ],
      "wer": 0.01180607887465461,
      "cer": 0.004210557638516931,
      "mer": 0.01179817588486319,
      "wil": 0.020508167521017695,
      "wip": 0.9794918324789823,
      "exact_match_rate": 0.9034926470588235,
      "utterance_mean_wer": 0.011642725636291814,
      "utterance_mean_cer": 0.004352088341262807,
      "rtf": 6.227715330972293,
      "latency_p50_s": 21.817871015518904,
      "latency_p95_s": 32.1434196209535,
      "peak_vram_mib": 15765.91650390625,
      "audio_seconds": 3808.4556916099773,
      "silence_ratio": 0.1599482895773222,
      "clipping_ratio": 0.0,
      "rms_dbfs": -21.204251726272272,
      "protocol": "Original provider configuration"
    },
    {
      "model": "chatterbox",
      "name": "Chatterbox",
      "checkpoint": "ResembleAI/chatterbox",
      "revision": "5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.010239624689752567,
        0.014502005846153599
      ],
      "cer_ci95": [
        0.002878993025473877,
        0.004195101111888829
      ],
      "wer": 0.012308465209746294,
      "cer": 0.0035261560435636496,
      "mer": 0.012300225922516944,
      "wil": 0.022405080632642016,
      "wip": 0.977594919367358,
      "exact_match_rate": 0.8851102941176471,
      "utterance_mean_wer": 0.011868301735029677,
      "utterance_mean_cer": 0.0034512612340441458,
      "rtf": 0.8215815223732541,
      "latency_p50_s": 2.598566422238946,
      "latency_p95_s": 3.5963106857612717,
      "peak_vram_mib": 3243.5419921875,
      "audio_seconds": 3491.6,
      "silence_ratio": 0.14410049492526528,
      "clipping_ratio": 3.745316243777417e-06,
      "rms_dbfs": -17.46151259613592,
      "protocol": "Original provider configuration"
    },
    {
      "model": "mosstts",
      "name": "MossTTS",
      "checkpoint": "OpenMOSS-Team/MOSS-TTS-v1.5",
      "revision": "cdd3b911b1585e3f2dbc7775ef10f9926f58850a",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.009509180623730396,
        0.015755375124648537
      ],
      "cer_ci95": [
        0.002815820019341391,
        0.006384203524120753
      ],
      "wer": 0.012308465209746294,
      "cer": 0.00431470570731417,
      "mer": 0.012298167823977245,
      "wil": 0.021827881301883956,
      "wip": 0.978172118698116,
      "exact_match_rate": 0.9080882352941176,
      "utterance_mean_wer": 0.012728509235862176,
      "utterance_mean_cer": 0.004697708975017243,
      "rtf": 1.389709319705265,
      "latency_p50_s": 7.407788587734103,
      "latency_p95_s": 10.18555185869335,
      "peak_vram_mib": 25009.28662109375,
      "audio_seconds": 5986.0,
      "silence_ratio": 0.43571122588738487,
      "clipping_ratio": 4.602443648000904e-06,
      "rms_dbfs": -26.128175142768608,
      "protocol": "Original provider configuration"
    },
    {
      "model": "outetts",
      "name": "OuteTTS",
      "checkpoint": "OuteAI/Llama-OuteTTS-1.0-1B",
      "revision": "f445d0d301c7fc9e47fd545eb9f9d2912afe7a77",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.010140987840998449,
        0.014841937026503233
      ],
      "cer_ci95": [
        0.003199460828193918,
        0.0057059813442127735
      ],
      "wer": 0.012392196265594908,
      "cer": 0.00437421888948402,
      "mer": 0.012376651613982272,
      "wil": 0.0210828903062108,
      "wip": 0.9789171096937892,
      "exact_match_rate": 0.8933823529411765,
      "utterance_mean_wer": 0.01187204084722467,
      "utterance_mean_cer": 0.004199895383034018,
      "rtf": 3.164612982009557,
      "latency_p50_s": 14.944048950448632,
      "latency_p95_s": 20.17052680440247,
      "peak_vram_mib": 4373.70263671875,
      "audio_seconds": 5142.874,
      "silence_ratio": 0.3346259142214407,
      "clipping_ratio": 0.0,
      "rms_dbfs": -19.94094323546973,
      "protocol": "Original provider configuration"
    },
    {
      "model": "vibevoice",
      "name": "VibeVoice",
      "checkpoint": "microsoft/VibeVoice-Realtime-0.5B",
      "revision": "6bce5f06044837fe6d2c5d7a71a84f0416bd57e4",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.010848131126616388,
        0.015379696839116497
      ],
      "cer_ci95": [
        0.003391400828390384,
        0.00499105574195605
      ],
      "wer": 0.013145775768232438,
      "cer": 0.004151044456347081,
      "mer": 0.013129285833751463,
      "wil": 0.023229322079901915,
      "wip": 0.9767706779200981,
      "exact_match_rate": 0.8823529411764706,
      "utterance_mean_wer": 0.012995374453257238,
      "utterance_mean_cer": 0.004069196706291303,
      "rtf": 0.6212332774264896,
      "latency_p50_s": 2.793820545077324,
      "latency_p95_s": 4.0911998309195035,
      "peak_vram_mib": 2034.2626953125,
      "audio_seconds": 4941.2,
      "silence_ratio": 0.31937791229579665,
      "clipping_ratio": 7.978451797385622e-09,
      "rms_dbfs": -26.039530312116856,
      "protocol": "Original provider configuration"
    },
    {
      "model": "qwen3tts",
      "name": "Qwen3TTS",
      "checkpoint": "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
      "revision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.011347743257994994,
        0.016629512899164478
      ],
      "cer_ci95": [
        0.00341322717010673,
        0.005637916399275065
      ],
      "wer": 0.013815624215021352,
      "cer": 0.0043890971850264835,
      "mer": 0.013802911159444537,
      "wil": 0.02513313514949267,
      "wip": 0.9748668648505073,
      "exact_match_rate": 0.8832720588235294,
      "utterance_mean_wer": 0.013853437463272021,
      "utterance_mean_cer": 0.004452845678980064,
      "rtf": 1.7696765178685818,
      "latency_p50_s": 8.603621456772089,
      "latency_p95_s": 13.723505410738287,
      "peak_vram_mib": 4351.36962890625,
      "audio_seconds": 5396.08,
      "silence_ratio": 0.2114373210618118,
      "clipping_ratio": 0.0,
      "rms_dbfs": -21.57469666576863,
      "protocol": "Original provider configuration"
    },
    {
      "model": "neutts",
      "name": "NeuTTS",
      "checkpoint": "neuphonic/neutts-2e",
      "revision": "412aaab11c6b727c6c0fe2552db305c234e568da",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.011031144307612399,
        0.019876654385442543
      ],
      "cer_ci95": [
        0.003660288780343977,
        0.010057701099493685
      ],
      "wer": 0.014736665829356108,
      "cer": 0.006204249241206927,
      "mer": 0.014692378328741965,
      "wil": 0.02428765130115995,
      "wip": 0.97571234869884,
      "exact_match_rate": 0.8887867647058824,
      "utterance_mean_wer": 0.014267839703068943,
      "utterance_mean_cer": 0.00575985081832371,
      "rtf": 1.7043660177076607,
      "latency_p50_s": 6.949139470234513,
      "latency_p95_s": 9.845434662140903,
      "peak_vram_mib": 3476.28369140625,
      "audio_seconds": 4472.06,
      "silence_ratio": 0.27913515952061047,
      "clipping_ratio": 4.345555472947041e-06,
      "rms_dbfs": -19.889727299836053,
      "protocol": "Original provider configuration"
    },
    {
      "model": "cosyvoice",
      "name": "CosyVoice 3 · 0.5B · 2512",
      "checkpoint": "artifacts/cosyvoice3",
      "revision": null,
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.014638886696980343,
        0.02044482781056171
      ],
      "cer_ci95": [
        0.005187774747425175,
        0.007428619023023338
      ],
      "wer": 0.017416059616511763,
      "cer": 0.0062340058322918526,
      "mer": 0.017391304347826087,
      "wil": 0.031161646058276316,
      "wip": 0.9688383539417237,
      "exact_match_rate": 0.8547794117647058,
      "utterance_mean_wer": 0.01791734740953491,
      "utterance_mean_cer": 0.006269259896464724,
      "rtf": 0.7025528225695982,
      "latency_p50_s": 2.636332143098116,
      "latency_p95_s": 3.7965015506371853,
      "peak_vram_mib": 3429.84326171875,
      "audio_seconds": 4108.52,
      "silence_ratio": 0.17668273965805056,
      "clipping_ratio": 0.0,
      "rms_dbfs": -18.83834829603301,
      "quality_review_reason": "Full 1088-text resynthesis verified after publisher-aligned HiFT repair.",
      "quality_audit": "cosyvoice-correction.html",
      "upstream_checkpoint": "FunAudioLLM/Fun-CosyVoice3-0.5B-2512",
      "upstream_revision": "29e01c4e8d000f4bcd70751be16fa94bf3d85a18",
      "version_label": "CosyVoice 3, 0.5B, December 2025 base checkpoint (llm.pt; not RL)",
      "score_status": "corrected_full_run",
      "correction": {
        "version": "cosyvoice-v3-hift-20260915",
        "verified_at": "2026-09-15T15:53:32.639125+00:00",
        "report": "cosyvoice-correction.html",
        "verification": "reports/cosyvoice-correction-2026-09-15/verification.json",
        "original_wer": 0.13815624215021352,
        "original_cer": 0.07709932750104148,
        "original_score_status": "invalidated_by_implementation_bug",
        "original_dataset_revision": "7d02b49322a9731ea9b1cc6279ef3f7a80949ae8",
        "original_audio_archive": "audio_shards/cosyvoice.tar",
        "original_records": "reports/cosyvoice-correction-2026-09-15/original-model-records.json",
        "corrected_dataset_revision": "56c87034f5ef6a4b7bcfe403390ed86a0fc36611",
        "corrected_audio_archive": "corrections/cosyvoice-v3-hift-20260915/cosyvoice.tar"
      },
      "protocol": "Original provider configuration"
    },
    {
      "model": "higgstts",
      "name": "HiggsTTS",
      "checkpoint": "bosonai/higgs-tts-2-3b-base",
      "revision": "d80c511612b3040ff2877ce3d408747df1739f11",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.014684263569190567,
        0.02052255964339718
      ],
      "cer_ci95": [
        0.005334938979585008,
        0.008322318642279253
      ],
      "wer": 0.017416059616511763,
      "cer": 0.006710111289650658,
      "mer": 0.01738113144480655,
      "wil": 0.02976327841893811,
      "wip": 0.9702367215810619,
      "exact_match_rate": 0.859375,
      "utterance_mean_wer": 0.017204365538464803,
      "utterance_mean_cer": 0.006782952438796437,
      "rtf": 1.171588010392337,
      "latency_p50_s": 4.582375276833773,
      "latency_p95_s": 7.186718510463833,
      "peak_vram_mib": 11947.0888671875,
      "audio_seconds": 4356.6,
      "silence_ratio": 0.2434151965391991,
      "clipping_ratio": 2.2705394611234025e-06,
      "rms_dbfs": -21.519289781106792,
      "protocol": "Original provider configuration"
    },
    {
      "model": "inflecttts",
      "name": "InflectTTS",
      "checkpoint": "owensong/Inflect-Micro-v2",
      "revision": "132edf8700da8e0d3fa81cdc0fd8844926fc6582",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.014729798756944202,
        0.020587894869992844
      ],
      "cer_ci95": [
        0.004852113065057902,
        0.006962616006936535
      ],
      "wer": 0.017583521728208994,
      "cer": 0.0058918050348152114,
      "mer": 0.017562933846282512,
      "wil": 0.03222911364088443,
      "wip": 0.9677708863591156,
      "exact_match_rate": 0.8566176470588235,
      "utterance_mean_wer": 0.017602403723727254,
      "utterance_mean_cer": 0.006094378628546624,
      "rtf": 0.014407729416794657,
      "latency_p50_s": 0.060717109590768814,
      "latency_p95_s": 0.07416072003543377,
      "peak_vram_mib": 174.49169921875,
      "audio_seconds": 4074.2906666666668,
      "silence_ratio": 0.23552280891102917,
      "clipping_ratio": 0.0,
      "rms_dbfs": -23.741855445278556,
      "protocol": "Original provider configuration"
    },
    {
      "model": "orpheustts",
      "name": "OrpheusTTS",
      "checkpoint": "canopylabs/orpheus-3b-0.1-ft",
      "revision": "4206a56e5a68cf6cf96900a8a78acd3370c02eb6",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.016988210857373835,
        0.023450519014427323
      ],
      "cer_ci95": [
        0.006083833741794046,
        0.009892766606762127
      ],
      "wer": 0.020011722347818808,
      "cer": 0.007692078795453193,
      "mer": 0.019973257563095436,
      "wil": 0.03503674213176289,
      "wip": 0.9649632578682371,
      "exact_match_rate": 0.8373161764705882,
      "utterance_mean_wer": 0.020058611476350446,
      "utterance_mean_cer": 0.008047379814062864,
      "rtf": 2.5830602675349486,
      "latency_p50_s": 11.744120076298714,
      "latency_p95_s": 17.74872815031558,
      "peak_vram_mib": 6409.50244140625,
      "audio_seconds": 5113.685333333333,
      "silence_ratio": 0.40942533109258156,
      "clipping_ratio": 0.0,
      "rms_dbfs": -27.90337897777544,
      "protocol": "Original provider configuration"
    },
    {
      "model": "zonos",
      "name": "Zonos",
      "checkpoint": "Zyphra/Zonos-v0.1-transformer",
      "revision": "9d8331fc49cb5ba8aad2bb56cafd809c66598f4e",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.02007390075144153,
        0.026153331156134465
      ],
      "cer_ci95": [
        0.008174096745526271,
        0.01130165207207019
      ],
      "wer": 0.02302604035836892,
      "cer": 0.0096411355115158,
      "mer": 0.022951093306626605,
      "wil": 0.037599506467079435,
      "wip": 0.9624004935329206,
      "exact_match_rate": 0.8023897058823529,
      "utterance_mean_wer": 0.023355699436872036,
      "utterance_mean_cer": 0.010020021987216247,
      "rtf": 1.7680858104657158,
      "latency_p50_s": 6.9678456503897905,
      "latency_p95_s": 10.227166134119033,
      "peak_vram_mib": 4144.82080078125,
      "audio_seconds": 4331.996009070294,
      "silence_ratio": 0.26132722105228,
      "clipping_ratio": 1.1080583572800664e-07,
      "rms_dbfs": -23.673586565392633,
      "protocol": "Original provider configuration"
    },
    {
      "model": "gptsovits",
      "name": "GPTSoVITS",
      "checkpoint": "lj1995/GPT-SoVITS",
      "revision": "336b2ec4e8d4ac74740798dd40af44e74659ecaf",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.022651802065312383,
        0.030441862466967787
      ],
      "cer_ci95": [
        0.007285552893791007,
        0.010238613992923822
      ],
      "wer": 0.02637528259231349,
      "cer": 0.008703802892340653,
      "mer": 0.02626532143750521,
      "wil": 0.04524414603704274,
      "wip": 0.9547558539629573,
      "exact_match_rate": 0.8226102941176471,
      "utterance_mean_wer": 0.02771551747884468,
      "utterance_mean_cer": 0.009832378998327021,
      "rtf": 0.3222476706943061,
      "latency_p50_s": 1.066016722470522,
      "latency_p95_s": 1.7044374926015722,
      "peak_vram_mib": 764.4111328125,
      "audio_seconds": 3679.6,
      "silence_ratio": 0.15031361795966175,
      "clipping_ratio": 0.0,
      "rms_dbfs": -21.53940639823217,
      "protocol": "Original provider configuration"
    },
    {
      "model": "melotts",
      "name": "MeloTTS",
      "checkpoint": "EN",
      "revision": null,
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.02316930251848474,
        0.029679677277896818
      ],
      "cer_ci95": [
        0.009173332758245023,
        0.012222008404797419
      ],
      "wer": 0.02637528259231349,
      "cer": 0.01065285960840326,
      "mer": 0.026328986960882646,
      "wil": 0.048229403790975844,
      "wip": 0.9517705962090242,
      "exact_match_rate": 0.7766544117647058,
      "utterance_mean_wer": 0.02676533790963938,
      "utterance_mean_cer": 0.01089498176697487,
      "rtf": 0.019975049763602603,
      "latency_p50_s": 0.1015011165291071,
      "latency_p95_s": 0.11808984335511923,
      "peak_vram_mib": 506.787109375,
      "audio_seconds": 4986.566530612245,
      "silence_ratio": 0.3701696886986826,
      "clipping_ratio": 0.0,
      "rms_dbfs": -26.93682087807902,
      "protocol": "Original provider configuration"
    },
    {
      "model": "parlertts",
      "name": "ParlerTTS",
      "checkpoint": "parler-tts/parler-tts-mini-v1",
      "revision": "0392b9451a601e528fd863bbb0598431fee810d9",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.0227290026732326,
        0.03095461490670336
      ],
      "cer_ci95": [
        0.012000777770688552,
        0.017373967449886223
      ],
      "wer": 0.026542744704010718,
      "cer": 0.0144617032672737,
      "mer": 0.026438698915763136,
      "wil": 0.04502352105370688,
      "wip": 0.9549764789462931,
      "exact_match_rate": 0.8079044117647058,
      "utterance_mean_wer": 0.027006843773840338,
      "utterance_mean_cer": 0.014403632229321195,
      "rtf": 1.8774146532842886,
      "latency_p50_s": 6.555586036294699,
      "latency_p95_s": 11.28402764480561,
      "peak_vram_mib": 4507.4521484375,
      "audio_seconds": 4018.3873015873014,
      "silence_ratio": 0.1515094306108054,
      "clipping_ratio": 7.417982043023822e-09,
      "rms_dbfs": -18.799491891228392,
      "protocol": "Original provider configuration"
    },
    {
      "model": "csm",
      "name": "CSM",
      "checkpoint": "sesame/csm-1b",
      "revision": "c92a71e1c419772e25be7dc14d952c2521a740ab",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.02256984690648647,
        0.03114456456487128
      ],
      "cer_ci95": [
        0.012244555826613683,
        0.017740773268843477
      ],
      "wer": 0.026710206815707946,
      "cer": 0.014878295542462656,
      "mer": 0.026462048942347573,
      "wil": 0.04070971618598751,
      "wip": 0.9592902838140125,
      "exact_match_rate": 0.8088235294117647,
      "utterance_mean_wer": 0.027264451475091614,
      "utterance_mean_cer": 0.015297082046186216,
      "rtf": 2.2531292765703044,
      "latency_p50_s": 8.784898072481155,
      "latency_p95_s": 15.126542358100414,
      "peak_vram_mib": 4330.97998046875,
      "audio_seconds": 4515.44,
      "silence_ratio": 0.3169689076414828,
      "clipping_ratio": 3.525773558828079e-05,
      "rms_dbfs": -26.242882987863332,
      "protocol": "Original provider configuration"
    },
    {
      "model": "voxcpm",
      "name": "VoxCPM",
      "checkpoint": "openbmb/VoxCPM2",
      "revision": "bffb3df5a29440629464e5e839f4d214c8714c3d",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.027210668207721767,
        0.04040970866798127
      ],
      "cer_ci95": [
        0.020480165206463754,
        0.03256538243822571
      ],
      "wer": 0.0334924223394457,
      "cer": 0.02599238231268226,
      "mer": 0.032706459525756335,
      "wil": 0.04076054667222251,
      "wip": 0.9592394533277775,
      "exact_match_rate": 0.8363970588235294,
      "utterance_mean_wer": 0.04057948791404673,
      "utterance_mean_cer": 0.03500154595326482,
      "rtf": 0.8742893887299773,
      "latency_p50_s": 3.0798537097871304,
      "latency_p95_s": 4.482512011565265,
      "peak_vram_mib": 5568.4033203125,
      "audio_seconds": 3915.68,
      "silence_ratio": 0.25207125031657623,
      "clipping_ratio": 0.0,
      "rms_dbfs": -19.995685779135187,
      "protocol": "Original provider configuration"
    },
    {
      "model": "openvoice",
      "name": "OpenVoice",
      "checkpoint": "myshell-ai/OpenVoiceV2",
      "revision": "f36e7edfe1684461a8343844af60babc2efbb727",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.03259408128993721,
        0.0412924532698379
      ],
      "cer_ci95": [
        0.013940913785427172,
        0.017979386205271203
      ],
      "wer": 0.03684166457339027,
      "cer": 0.015904897934892578,
      "mer": 0.03674014696058784,
      "wil": 0.06683946355663117,
      "wip": 0.9331605364433688,
      "exact_match_rate": 0.7205882352941176,
      "utterance_mean_wer": 0.039472592478647844,
      "utterance_mean_cer": 0.01727688301463179,
      "rtf": 0.041566943720588286,
      "latency_p50_s": 0.18259630538523197,
      "latency_p95_s": 0.22698942646384238,
      "peak_vram_mib": 752.9150390625,
      "audio_seconds": 4986.566530612245,
      "silence_ratio": 0.3955355348300485,
      "clipping_ratio": 0.0,
      "rms_dbfs": -27.237915118797478,
      "protocol": "Original provider configuration"
    },
    {
      "model": "vits",
      "name": "Vits",
      "checkpoint": "facebook/mms-tts-eng",
      "revision": "c71de0fe7204c83f1c10820a7d696d0b450048ba",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.0568077628669862,
        0.06791827951441402
      ],
      "cer_ci95": [
        0.024251829595567223,
        0.02939332583322282
      ],
      "wer": 0.062295905551369005,
      "cer": 0.02673629708980539,
      "mer": 0.061881393994843215,
      "wil": 0.10974369514421811,
      "wip": 0.8902563048557819,
      "exact_match_rate": 0.6084558823529411,
      "utterance_mean_wer": 0.0668014646232554,
      "utterance_mean_cer": 0.029005364603199517,
      "rtf": 0.036346979152729886,
      "latency_p50_s": 0.1107128206640482,
      "latency_p95_s": 0.3603286813944578,
      "peak_vram_mib": 161.82177734375,
      "audio_seconds": 4713.184,
      "silence_ratio": 0.24361948208129203,
      "clipping_ratio": 0.0,
      "rms_dbfs": -18.456575731465676,
      "protocol": "Original provider configuration"
    },
    {
      "model": "bark",
      "name": "Bark",
      "checkpoint": "suno/bark-small",
      "revision": "1dbd7a128513b8ae4a4e2130fed57b7ac9da5bcd",
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.06542025882580918,
        0.09126204860193128
      ],
      "cer_ci95": [
        0.041101972027671005,
        0.06016230488018854
      ],
      "wer": 0.07736749560411957,
      "cer": 0.049648872225197885,
      "mer": 0.07522592200602458,
      "wil": 0.11118412348963358,
      "wip": 0.8888158765103664,
      "exact_match_rate": 0.6461397058823529,
      "utterance_mean_wer": 0.08971144156575432,
      "utterance_mean_cer": 0.06200807096575261,
      "rtf": 1.3509386013124312,
      "latency_p50_s": 7.324050920084119,
      "latency_p95_s": 17.736357633210712,
      "peak_vram_mib": 1866.55224609375,
      "audio_seconds": 6735.826666666667,
      "silence_ratio": 0.3688515404183891,
      "clipping_ratio": 4.311951581619864e-05,
      "rms_dbfs": -25.216558823343817,
      "protocol": "Original provider configuration"
    },
    {
      "model": "conversationtts",
      "name": "ConversationTTS",
      "checkpoint": "artifacts/conversationtts/model.safetensors",
      "revision": null,
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.06827334380474649,
        0.10941677412955063
      ],
      "cer_ci95": [
        0.05507506031858041,
        0.11838151593873322
      ],
      "wer": 0.08716402913840743,
      "cer": 0.0824703921918705,
      "mer": 0.08171756024805715,
      "wil": 0.0985030718625245,
      "wip": 0.9014969281374755,
      "exact_match_rate": 0.765625,
      "utterance_mean_wer": 0.09572835930129399,
      "utterance_mean_cer": 0.10447683402668193,
      "rtf": 2.6722347402449587,
      "latency_p50_s": 11.374369565397501,
      "latency_p95_s": 21.11884947624052,
      "peak_vram_mib": 4320.94140625,
      "audio_seconds": 5198.24,
      "silence_ratio": 0.22706196691592928,
      "clipping_ratio": 0.0010200252838765175,
      "rms_dbfs": -16.481099157390705,
      "protocol": "Original provider configuration"
    },
    {
      "model": "vui",
      "name": "Vui",
      "checkpoint": "vui-abraham-100m.pt",
      "revision": null,
      "scored": 1088,
      "quality_review": false,
      "empty_asr_transcripts": 0,
      "wer_ci95": [
        0.10961879047597937,
        0.13331448614544683
      ],
      "cer_ci95": [
        0.07456182716011253,
        0.09409126504207617
      ],
      "wer": 0.12149376203633928,
      "cer": 0.08397310004165923,
      "mer": 0.11439608956165248,
      "wil": 0.15491850701861398,
      "wip": 0.845081492981386,
      "exact_match_rate": 0.4963235294117647,
      "utterance_mean_wer": 0.1461771074776305,
      "utterance_mean_cer": 0.11014174174451084,
      "rtf": 0.4120764092569282,
      "latency_p50_s": 1.9950699619948864,
      "latency_p95_s": 2.9318609584122894,
      "peak_vram_mib": 3020.765625,
      "audio_seconds": 5821.52126984127,
      "silence_ratio": 0.36983417003042174,
      "clipping_ratio": 1.0284511102981117e-05,
      "rms_dbfs": -16.98248013674346,
      "protocol": "Original provider configuration"
    },
    {
      "model": "dia",
      "name": "Dia",
      "checkpoint": "nari-labs/Dia-1.6B-0626",
      "revision": "ef2795fcc29c5abe6ffc91fd33808588b49bbc66",
      "scored": 1088,
      "quality_review": true,
      "empty_asr_transcripts": 183,
      "wer_ci95": [
        0.5772596992606132,
        0.7770414764299733
      ],
      "cer_ci95": [
        0.5068358244822976,
        0.6812881581151855
      ],
      "wer": 0.673532613246253,
      "cer": 0.5928256858894245,
      "mer": 0.5234934270467265,
      "wil": 0.6199658900150475,
      "wip": 0.3800341099849525,
      "exact_match_rate": 0.4522058823529412,
      "utterance_mean_wer": 0.7629098579777105,
      "utterance_mean_cer": 0.6984215456512638,
      "rtf": 3.1889203149742467,
      "latency_p50_s": 35.84547644108534,
      "latency_p95_s": 95.22943165823817,
      "peak_vram_mib": 9100.61865234375,
      "audio_seconds": 15192.758276643992,
      "silence_ratio": 0.19584809755048657,
      "clipping_ratio": 4.931407476340513e-08,
      "rms_dbfs": -22.07775442813294,
      "quality_review_reason": "High WER remains under review. Three native/Transformers generation comparisons and independent DAC decoding did not establish a port defect. Empty transcripts and ASR repetition affect scores; diagnostic VAD does not restore missing speech.",
      "quality_audit": "quality-audit.html",
      "protocol": "Original provider configuration"
    },
    {
      "attempted": 1088,
      "generated": 1088,
      "scored": 1088,
      "generation_failure_rate": 0.0,
      "latency_p50_s": 2.898507172241807,
      "latency_p95_s": 4.304634319432079,
      "rtf": 0.7075289849843672,
      "peak_vram_mib": 5139.884765625,
      "audio_seconds": 4548.18,
      "clipping_ratio": 1.2887815819270687e-07,
      "silence_ratio": 0.30699522734627727,
      "rms_dbfs": -21.0199655988594,
      "normalization_id": "whisper_english",
      "wer": 0.011554885707108766,
      "cer": 0.004151044456347081,
      "mer": 0.011548117154811716,
      "wil": 0.019847561989309592,
      "wip": 0.9801524380106904,
      "word_hits": 11812,
      "word_substitutions": 100,
      "word_deletions": 31,
      "word_insertions": 7,
      "reference_words": 11943,
      "char_hits": 66984,
      "char_substitutions": 72,
      "char_deletions": 156,
      "char_insertions": 51,
      "reference_chars": 67212,
      "exact_match_rate": 0.9090073529411765,
      "raw": {
        "wer": 0.05199864498644986,
        "cer": 0.013326966487237368,
        "mer": 0.05198543730420794,
        "wil": 0.09714150184125148,
        "wip": 0.9028584981587485,
        "word_hits": 11197,
        "word_substitutions": 560,
        "word_deletions": 51,
        "word_insertions": 3,
        "reference_words": 11808,
        "char_hits": 68314,
        "char_substitutions": 361,
        "char_deletions": 433,
        "char_insertions": 127,
        "reference_chars": 69108,
        "exact_match_rate": 0.6525735294117647
      },
      "orthographic": {
        "wer": 0.013626745662293695,
        "cer": 0.006061776635416821,
        "mer": 0.013624439367013625,
        "wil": 0.0241734777985686,
        "wip": 0.9758265222014314,
        "word_hits": 11656,
        "word_substitutions": 126,
        "word_deletions": 33,
        "word_insertions": 2,
        "reference_words": 11815,
        "char_hits": 66935,
        "char_substitutions": 105,
        "char_deletions": 267,
        "char_insertions": 36,
        "reference_chars": 67307,
        "exact_match_rate": 0.8878676470588235
      },
      "wer_ci95": [
        0.009040538395502281,
        0.014536809274260744
      ],
      "cer_ci95": [
        0.002496536485865647,
        0.006234852324878584
      ],
      "unique_prompts_scored": 1088,
      "by_category": {
        "read_speech": {
          "wer": 0.011554885707108766,
          "cer": 0.004151044456347081,
          "mer": 0.011548117154811716,
          "wil": 0.019847561989309592,
          "wip": 0.9801524380106904,
          "word_hits": 11812,
          "word_substitutions": 100,
          "word_deletions": 31,
          "word_insertions": 7,
          "reference_words": 11943,
          "char_hits": 66984,
          "char_substitutions": 72,
          "char_deletions": 156,
          "char_insertions": 51,
          "reference_chars": 67212,
          "exact_match_rate": 0.9090073529411765
        }
      },
      "utterance_mean_wer": 0.010836076709238474,
      "utterance_mean_cer": 0.0038127112921178003,
      "generation_failures": 0,
      "asr_scored": 1088,
      "failure_policy": "empty-output-deletions-v1",
      "model": "llasa-1b",
      "name": "Llasa-1B · ref",
      "checkpoint": "HKUSTAudio/Llasa-1B",
      "revision": "cef8ce6a6eea3b593c5b28868d2391ec15eb9058",
      "protocol": "Fixed reference · independent implementation",
      "experiment": "reference-conditioned",
      "records_path": "data/variants/full/llasa-1b.json"
    },
    {
      "attempted": 1088,
      "generated": 1088,
      "scored": 1088,
      "generation_failure_rate": 0.0,
      "latency_p50_s": 4.488604974001646,
      "latency_p95_s": 6.565801392495629,
      "rtf": 1.1598871871001017,
      "peak_vram_mib": 9036.13232421875,
      "audio_seconds": 4301.1,
      "clipping_ratio": 3.6179961565716633e-07,
      "silence_ratio": 0.2722888598285631,
      "rms_dbfs": -20.879893134348087,
      "normalization_id": "whisper_english",
      "wer": 0.009377878255044795,
      "cer": 0.0025739451288460395,
      "mer": 0.009373953799799129,
      "wil": 0.017097483085435683,
      "wip": 0.9829025169145643,
      "word_hits": 11836,
      "word_substitutions": 93,
      "word_deletions": 14,
      "word_insertions": 5,
      "reference_words": 11943,
      "char_hits": 67091,
      "char_substitutions": 65,
      "char_deletions": 56,
      "char_insertions": 52,
      "reference_chars": 67212,
      "exact_match_rate": 0.9154411764705882,
      "raw": {
        "wer": 0.051490514905149054,
        "cer": 0.011894426115645076,
        "mer": 0.05147743628820591,
        "wil": 0.09755492849651703,
        "wip": 0.902445071503483,
        "word_hits": 11203,
        "word_substitutions": 572,
        "word_deletions": 33,
        "word_insertions": 3,
        "reference_words": 11808,
        "char_hits": 68410,
        "char_substitutions": 357,
        "char_deletions": 341,
        "char_insertions": 124,
        "reference_chars": 69108,
        "exact_match_rate": 0.6452205882352942
      },
      "orthographic": {
        "wer": 0.011341515023275498,
        "cer": 0.004397759519812204,
        "mer": 0.01133959549801134,
        "wil": 0.021142625359624412,
        "wip": 0.9788573746403756,
        "word_hits": 11683,
        "word_substitutions": 117,
        "word_deletions": 15,
        "word_insertions": 2,
        "reference_words": 11815,
        "char_hits": 67053,
        "char_substitutions": 96,
        "char_deletions": 158,
        "char_insertions": 42,
        "reference_chars": 67307,
        "exact_match_rate": 0.8961397058823529
      },
      "wer_ci95": [
        0.0075335024885588775,
        0.011436463847120435
      ],
      "cer_ci95": [
        0.002015836833011332,
        0.003193140342505806
      ],
      "unique_prompts_scored": 1088,
      "by_category": {
        "read_speech": {
          "wer": 0.009377878255044795,
          "cer": 0.0025739451288460395,
          "mer": 0.009373953799799129,
          "wil": 0.017097483085435683,
          "wip": 0.9829025169145643,
          "word_hits": 11836,
          "word_substitutions": 93,
          "word_deletions": 14,
          "word_insertions": 5,
          "reference_words": 11943,
          "char_hits": 67091,
          "char_substitutions": 65,
          "char_deletions": 56,
          "char_insertions": 52,
          "reference_chars": 67212,
          "exact_match_rate": 0.9154411764705882
        }
      },
      "utterance_mean_wer": 0.009197328406887232,
      "utterance_mean_cer": 0.0025238450023400924,
      "generation_failures": 0,
      "asr_scored": 1088,
      "failure_policy": "empty-output-deletions-v1",
      "model": "llasa-3b",
      "name": "Llasa-3B · ref",
      "checkpoint": "HKUSTAudio/Llasa-3B",
      "revision": "740b08c956fdde4a5a33f2caffbf004ff76e2748",
      "protocol": "Fixed reference · independent implementation",
      "experiment": "reference-conditioned",
      "records_path": "data/variants/full/llasa-3b.json"
    },
    {
      "attempted": 1088,
      "generated": 1031,
      "scored": 1088,
      "generation_failure_rate": 0.05238970588235292,
      "latency_p50_s": 5.182631328701973,
      "latency_p95_s": 7.3562819845974445,
      "rtf": 1.3018688388035875,
      "peak_vram_mib": 18886.0048828125,
      "audio_seconds": 4135.14,
      "clipping_ratio": 3.9953714786206674e-07,
      "silence_ratio": 0.2878392129514611,
      "rms_dbfs": -21.080921045435048,
      "normalization_id": "whisper_english",
      "wer": 0.05182952357029222,
      "cer": 0.04292388264000476,
      "mer": 0.05179482888461217,
      "wil": 0.05945953800730752,
      "wip": 0.9405404619926925,
      "word_hits": 11332,
      "word_substitutions": 92,
      "word_deletions": 519,
      "word_insertions": 8,
      "reference_words": 11943,
      "char_hits": 64376,
      "char_substitutions": 63,
      "char_deletions": 2773,
      "char_insertions": 49,
      "reference_chars": 67212,
      "exact_match_rate": 0.8630514705882353,
      "raw": {
        "wer": 0.09307249322493225,
        "cer": 0.05220813798691903,
        "mer": 0.09304097527937691,
        "wil": 0.13826195184662193,
        "wip": 0.8617380481533781,
        "word_hits": 10713,
        "word_substitutions": 562,
        "word_deletions": 533,
        "word_insertions": 4,
        "reference_words": 11808,
        "char_hits": 65625,
        "char_substitutions": 347,
        "char_deletions": 3136,
        "char_insertions": 125,
        "reference_chars": 69108,
        "exact_match_rate": 0.6075367647058824
      },
      "orthographic": {
        "wer": 0.05382987727465087,
        "cer": 0.04495817671267476,
        "mer": 0.053816212557116265,
        "wil": 0.0633760877576266,
        "wip": 0.9366239122423734,
        "word_hits": 11182,
        "word_substitutions": 114,
        "word_deletions": 519,
        "word_insertions": 3,
        "reference_words": 11815,
        "char_hits": 64316,
        "char_substitutions": 86,
        "char_deletions": 2905,
        "char_insertions": 35,
        "reference_chars": 67307,
        "exact_match_rate": 0.8455882352941176
      },
      "wer_ci95": [
        0.04124006661479701,
        0.06337072792519431
      ],
      "cer_ci95": [
        0.03248171656578204,
        0.053719000348595045
      ],
      "unique_prompts_scored": 1088,
      "by_category": {
        "read_speech": {
          "wer": 0.05182952357029222,
          "cer": 0.04292388264000476,
          "mer": 0.05179482888461217,
          "wil": 0.05945953800730752,
          "wip": 0.9405404619926925,
          "word_hits": 11332,
          "word_substitutions": 92,
          "word_deletions": 519,
          "word_insertions": 8,
          "reference_words": 11943,
          "char_hits": 64376,
          "char_substitutions": 63,
          "char_deletions": 2773,
          "char_insertions": 49,
          "reference_chars": 67212,
          "exact_match_rate": 0.8630514705882353
        }
      },
      "utterance_mean_wer": 0.06150821961943286,
      "utterance_mean_cer": 0.05468215056110834,
      "generation_failures": 57,
      "asr_scored": 1031,
      "failure_policy": "empty-output-deletions-v1",
      "successful_audio_only": {
        "attempted": 1031,
        "generated": 1031,
        "scored": 1031,
        "generation_failure_rate": 0.0,
        "latency_p50_s": 5.182631328701973,
        "latency_p95_s": 7.3562819845974445,
        "rtf": 1.3018688388035875,
        "peak_vram_mib": 18886.0048828125,
        "audio_seconds": 4135.14,
        "clipping_ratio": 3.9953714786206674e-07,
        "silence_ratio": 0.2878392129514611,
        "rms_dbfs": -21.080921045435048,
        "normalization_id": "whisper_english",
        "wer": 0.009966777408637873,
        "cer": 0.0025584568628667123,
        "mer": 0.009959811287786125,
        "wil": 0.01793366518808126,
        "wip": 0.9820663348119187,
        "word_hits": 11332,
        "word_substitutions": 92,
        "word_deletions": 14,
        "word_insertions": 8,
        "reference_words": 11438,
        "char_hits": 64376,
        "char_substitutions": 63,
        "char_deletions": 53,
        "char_insertions": 49,
        "reference_chars": 64492,
        "exact_match_rate": 0.9107662463627546,
        "raw": {
          "wer": 0.05238474471285727,
          "cer": 0.011648961854176727,
          "mer": 0.05236620964175144,
          "wil": 0.0996015509605267,
          "wip": 0.9003984490394733,
          "word_hits": 10713,
          "word_substitutions": 562,
          "word_deletions": 26,
          "word_insertions": 4,
          "reference_words": 11301,
          "char_hits": 65625,
          "char_substitutions": 347,
          "char_deletions": 300,
          "char_insertions": 125,
          "reference_chars": 66272,
          "exact_match_rate": 0.6411251212415131
        },
        "orthographic": {
          "wer": 0.011495269254576002,
          "cer": 0.0039667167671258345,
          "mer": 0.011492220650636492,
          "wil": 0.021468607025940223,
          "wip": 0.9785313929740598,
          "word_hits": 11182,
          "word_substitutions": 114,
          "word_deletions": 13,
          "word_insertions": 3,
          "reference_words": 11309,
          "char_hits": 64316,
          "char_substitutions": 86,
          "char_deletions": 135,
          "char_insertions": 35,
          "reference_chars": 64537,
          "exact_match_rate": 0.8923375363724539
        },
        "wer_ci95": [
          0.00804508819292334,
          0.01196427774151644
        ],
        "cer_ci95": [
          0.002022174986088534,
          0.0031449560486978313
        ],
        "unique_prompts_scored": 1031,
        "by_category": {
          "read_speech": {
            "wer": 0.009966777408637873,
            "cer": 0.0025584568628667123,
            "mer": 0.009959811287786125,
            "wil": 0.01793366518808126,
            "wip": 0.9820663348119187,
            "word_hits": 11332,
            "word_substitutions": 92,
            "word_deletions": 14,
            "word_insertions": 8,
            "reference_words": 11438,
            "char_hits": 64376,
            "char_substitutions": 63,
            "char_deletions": 53,
            "char_insertions": 49,
            "reference_chars": 64492,
            "exact_match_rate": 0.9107662463627546
          }
        },
        "utterance_mean_wer": 0.00962264107268957,
        "utterance_mean_cer": 0.0024191850732161804
      },
      "model": "llasa-8b",
      "name": "Llasa-8B · ref",
      "checkpoint": "HKUSTAudio/Llasa-8B",
      "revision": "2a8339e387348a5cad5160b978a6f364fa06ad74",
      "protocol": "Fixed reference · independent implementation",
      "experiment": "reference-conditioned",
      "records_path": "data/variants/full/llasa-8b.json"
    },
    {
      "attempted": 1088,
      "generated": 1088,
      "scored": 1088,
      "generation_failure_rate": 0.0,
      "latency_p50_s": 6.986981520429254,
      "latency_p95_s": 10.413049097731706,
      "rtf": 2.4360165730889434,
      "peak_vram_mib": 7114.560546875,
      "audio_seconds": 3183.2003628117914,
      "clipping_ratio": 4.622629233447537e-08,
      "silence_ratio": 0.15155201862902837,
      "rms_dbfs": -19.436149215137057,
      "normalization_id": "whisper_english",
      "wer": 0.049652516118228254,
      "cer": 0.043161935368684165,
      "mer": 0.04952396859863037,
      "wil": 0.07033722341818538,
      "wip": 0.9296627765818146,
      "word_hits": 11381,
      "word_substitutions": 254,
      "word_deletions": 308,
      "word_insertions": 31,
      "reference_words": 11943,
      "char_hits": 64483,
      "char_substitutions": 362,
      "char_deletions": 2367,
      "char_insertions": 172,
      "reference_chars": 67212,
      "exact_match_rate": 0.5882352941176471,
      "raw": {
        "wer": 0.10213414634146341,
        "cer": 0.05636105805406031,
        "mer": 0.10194420963651733,
        "wil": 0.16909589123707647,
        "wip": 0.8309041087629235,
        "word_hits": 10624,
        "word_substitutions": 858,
        "word_deletions": 326,
        "word_insertions": 22,
        "reference_words": 11808,
        "char_hits": 65451,
        "char_substitutions": 755,
        "char_deletions": 2902,
        "char_insertions": 238,
        "reference_chars": 69108,
        "exact_match_rate": 0.3786764705882353
      },
      "orthographic": {
        "wer": 0.051967837494710116,
        "cer": 0.04518103614780038,
        "mer": 0.05186686940361548,
        "wil": 0.07491215466858525,
        "wip": 0.9250878453314147,
        "word_hits": 11224,
        "word_substitutions": 279,
        "word_deletions": 312,
        "word_insertions": 23,
        "reference_words": 11815,
        "char_hits": 64422,
        "char_substitutions": 385,
        "char_deletions": 2500,
        "char_insertions": 156,
        "reference_chars": 67307,
        "exact_match_rate": 0.5680147058823529
      },
      "wer_ci95": [
        0.04563092458044647,
        0.05386343651680554
      ],
      "cer_ci95": [
        0.03943928765515252,
        0.04700508849635247
      ],
      "unique_prompts_scored": 1088,
      "by_category": {
        "read_speech": {
          "wer": 0.049652516118228254,
          "cer": 0.043161935368684165,
          "mer": 0.04952396859863037,
          "wil": 0.07033722341818538,
          "wip": 0.9296627765818146,
          "word_hits": 11381,
          "word_substitutions": 254,
          "word_deletions": 308,
          "word_insertions": 31,
          "reference_words": 11943,
          "char_hits": 64483,
          "char_substitutions": 362,
          "char_deletions": 2367,
          "char_insertions": 172,
          "reference_chars": 67212,
          "exact_match_rate": 0.5882352941176471
        }
      },
      "utterance_mean_wer": 0.05195949476387779,
      "utterance_mean_cer": 0.044414172358740525,
      "generation_failures": 0,
      "asr_scored": 1088,
      "failure_policy": "empty-output-deletions-v1",
      "model": "dia-1.6b-reference",
      "name": "Dia 1.6B · reference · ref",
      "checkpoint": "nari-labs/Dia-1.6B-0626",
      "revision": "ef2795fcc29c5abe6ffc91fd33808588b49bbc66",
      "protocol": "Fixed reference · independent implementation",
      "experiment": "reference-conditioned",
      "records_path": "data/variants/full/dia-1.6b-reference.json"
    },
    {
      "attempted": 1088,
      "generated": 1088,
      "scored": 1088,
      "generation_failure_rate": 0.0,
      "latency_p50_s": 11.62822395004332,
      "latency_p95_s": 12.75420665051788,
      "rtf": 2.7462189545183864,
      "peak_vram_mib": 5193.4755859375,
      "audio_seconds": 4612.72,
      "clipping_ratio": 0.0001029354660268676,
      "silence_ratio": 0.3528342140856518,
      "rms_dbfs": -20.42296298709414,
      "normalization_id": "whisper_english",
      "wer": 0.02754751737419409,
      "cer": 0.014193893947509374,
      "mer": 0.027467022875271332,
      "wil": 0.04958282078265752,
      "wip": 0.9504171792173425,
      "word_hits": 11649,
      "word_substitutions": 271,
      "word_deletions": 23,
      "word_insertions": 35,
      "reference_words": 11943,
      "char_hits": 66506,
      "char_substitutions": 454,
      "char_deletions": 252,
      "char_insertions": 248,
      "reference_chars": 67212,
      "exact_match_rate": 0.7757352941176471,
      "raw": {
        "wer": 0.06944444444444445,
        "cer": 0.023976963593216414,
        "mer": 0.06929772669652666,
        "wil": 0.12901508077342283,
        "wip": 0.8709849192265772,
        "word_hits": 11013,
        "word_substitutions": 755,
        "word_deletions": 40,
        "word_insertions": 25,
        "reference_words": 11808,
        "char_hits": 67762,
        "char_substitutions": 803,
        "char_deletions": 543,
        "char_insertions": 311,
        "reference_chars": 69108,
        "exact_match_rate": 0.5523897058823529
      },
      "orthographic": {
        "wer": 0.030215827338129497,
        "cer": 0.016253881468495105,
        "mer": 0.030136755022792505,
        "wil": 0.054902714969303434,
        "wip": 0.9450972850306966,
        "word_hits": 11489,
        "word_substitutions": 301,
        "word_deletions": 25,
        "word_insertions": 31,
        "reference_words": 11815,
        "char_hits": 66448,
        "char_substitutions": 489,
        "char_deletions": 370,
        "char_insertions": 235,
        "reference_chars": 67307,
        "exact_match_rate": 0.7564338235294118
      },
      "wer_ci95": [
        0.02424945783152844,
        0.0310445546903704
      ],
      "cer_ci95": [
        0.012211420128244528,
        0.016116875814609084
      ],
      "unique_prompts_scored": 1088,
      "by_category": {
        "read_speech": {
          "wer": 0.02754751737419409,
          "cer": 0.014193893947509374,
          "mer": 0.027467022875271332,
          "wil": 0.04958282078265752,
          "wip": 0.9504171792173425,
          "word_hits": 11649,
          "word_substitutions": 271,
          "word_deletions": 23,
          "word_insertions": 35,
          "reference_words": 11943,
          "char_hits": 66506,
          "char_substitutions": 454,
          "char_deletions": 252,
          "char_insertions": 248,
          "reference_chars": 67212,
          "exact_match_rate": 0.7757352941176471
        }
      },
      "utterance_mean_wer": 0.02924490179611486,
      "utterance_mean_cer": 0.014553274452331214,
      "generation_failures": 0,
      "asr_scored": 1088,
      "failure_policy": "empty-output-deletions-v1",
      "model": "dia2-1b",
      "name": "Dia2-1B · ref",
      "checkpoint": "nari-labs/Dia2-1B",
      "revision": "00042629c61c3268a6473552c911966ec7a5a450",
      "protocol": "Fixed reference · independent implementation",
      "experiment": "reference-conditioned",
      "records_path": "data/variants/full/dia2-1b.json"
    },
    {
      "attempted": 1088,
      "generated": 1088,
      "scored": 1088,
      "generation_failure_rate": 0.0,
      "latency_p50_s": 11.97853497043252,
      "latency_p95_s": 13.291872068122029,
      "rtf": 2.8539758534153523,
      "peak_vram_mib": 8386.26953125,
      "audio_seconds": 4579.44,
      "clipping_ratio": 0.00010766909432024053,
      "silence_ratio": 0.3425905363240454,
      "rms_dbfs": -20.067767057731597,
      "normalization_id": "whisper_english",
      "wer": 0.02168634346479109,
      "cer": 0.009536987442718562,
      "mer": 0.021648278167836845,
      "wil": 0.0396196850625794,
      "wip": 0.9603803149374206,
      "word_hits": 11705,
      "word_substitutions": 219,
      "word_deletions": 19,
      "word_insertions": 21,
      "reference_words": 11943,
      "char_hits": 66708,
      "char_substitutions": 307,
      "char_deletions": 197,
      "char_insertions": 137,
      "reference_chars": 67212,
      "exact_match_rate": 0.8143382352941176,
      "raw": {
        "wer": 0.06343157181571815,
        "cer": 0.018999247554552293,
        "mer": 0.06338326140306338,
        "wil": 0.1196966195372603,
        "wip": 0.8803033804627397,
        "word_hits": 11068,
        "word_substitutions": 708,
        "word_deletions": 32,
        "word_insertions": 9,
        "reference_words": 11808,
        "char_hits": 67992,
        "char_substitutions": 620,
        "char_deletions": 496,
        "char_insertions": 197,
        "reference_chars": 69108,
        "exact_match_rate": 0.578125
      },
      "orthographic": {
        "wer": 0.023614049936521372,
        "cer": 0.011202400938981087,
        "mer": 0.023590090470956285,
        "wil": 0.04411976874741241,
        "wip": 0.9558802312525876,
        "word_hits": 11548,
        "word_substitutions": 248,
        "word_deletions": 19,
        "word_insertions": 12,
        "reference_words": 11815,
        "char_hits": 66669,
        "char_substitutions": 332,
        "char_deletions": 306,
        "char_insertions": 116,
        "reference_chars": 67307,
        "exact_match_rate": 0.7950367647058824
      },
      "wer_ci95": [
        0.0188867689114121,
        0.02475265123752024
      ],
      "cer_ci95": [
        0.007964731585759912,
        0.01117886140291653
      ],
      "unique_prompts_scored": 1088,
      "by_category": {
        "read_speech": {
          "wer": 0.02168634346479109,
          "cer": 0.009536987442718562,
          "mer": 0.021648278167836845,
          "wil": 0.0396196850625794,
          "wip": 0.9603803149374206,
          "word_hits": 11705,
          "word_substitutions": 219,
          "word_deletions": 19,
          "word_insertions": 21,
          "reference_words": 11943,
          "char_hits": 66708,
          "char_substitutions": 307,
          "char_deletions": 197,
          "char_insertions": 137,
          "reference_chars": 67212,
          "exact_match_rate": 0.8143382352941176
        }
      },
      "utterance_mean_wer": 0.023010407984043822,
      "utterance_mean_cer": 0.010084375639320813,
      "generation_failures": 0,
      "asr_scored": 1088,
      "failure_policy": "empty-output-deletions-v1",
      "model": "dia2-2b",
      "name": "Dia2-2B · ref",
      "checkpoint": "nari-labs/Dia2-2B",
      "revision": "7abae125471a73b0fc6b9d413cb15f4ae1e771d8",
      "protocol": "Fixed reference · independent implementation",
      "experiment": "reference-conditioned",
      "records_path": "data/variants/full/dia2-2b.json"
    }
  ],
  "audio_bytes": 19539199408,
  "source_url": "https://github.com/BytedanceSpeech/seed-tts-eval",
  "audio_format": "indexed-wav-tar",
  "quality_audit": {
    "date": "2026-09-15",
    "report": "quality-audit.html",
    "original_scores_preserved": true,
    "cosyvoice_corrected_full_run": "completed",
    "corrected_full_score_available": true
  },
  "updated_at": "2026-09-15T15:56:52.325543+00:00",
  "corrections": [
    {
      "version": "cosyvoice-v3-hift-20260915",
      "verified_at": "2026-09-15T15:53:32.639125+00:00",
      "report": "cosyvoice-correction.html",
      "verification": "reports/cosyvoice-correction-2026-09-15/verification.json",
      "original_wer": 0.13815624215021352,
      "original_cer": 0.07709932750104148,
      "original_score_status": "invalidated_by_implementation_bug",
      "original_dataset_revision": "7d02b49322a9731ea9b1cc6279ef3f7a80949ae8",
      "original_audio_archive": "audio_shards/cosyvoice.tar",
      "original_records": "reports/cosyvoice-correction-2026-09-15/original-model-records.json",
      "corrected_dataset_revision": "56c87034f5ef6a4b7bcfe403390ed86a0fc36611",
      "corrected_audio_archive": "corrections/cosyvoice-v3-hift-20260915/cosyvoice.tar"
    }
  ],
  "base_snapshot_sha256": "0de7dd4c3af30aead1f97e51c00cd7970a81b62f2a1d52fe65cef35609690dcc",
  "variant_campaign": "dia-llasa-variants-20260915",
  "active_model_ids": [
    "llasa-1b",
    "llasa-3b",
    "llasa-8b",
    "dia-1.6b-reference",
    "dia2-1b",
    "dia2-2b"
  ],
  "excluded_base_model_ids": [
    "llasa"
  ]
}