{ "datasets": [ { "info": { "id": "vimedcss", "name": "ViMedCSS", "status": "live", "fullName": "Vietnamese Medical Code-Switching Speech", "tagline": "Code-switched medical speech from Vietnamese lectures — every utterance carries at least one English clinical term.", "source": "Vietnamese medical YouTube lectures", "hours": 34.57, "utterances": 16576, "csTerms": 889, "license": "CC-BY-4.0", "arxiv": "2602.12911", "hf": "tensorxt/ViMedCSS", "splits": { "train": { "hours": 24.31, "utts": 11833 }, "valid": { "hours": 3.56, "utts": 1714 }, "test": { "hours": 3.38, "utts": 1615 }, "hard": { "hours": 1.38, "utts": 658 } } }, "models": [ { "slug": "phowhisper-small-ag", "name": "PhoWhisper-small + Attention-Guide", "org": "ViMedCSS paper", "paramsB": 0.244, "license": "BSD-3-Clause", "category": "vn-open", "track": "C", "provenance": "paper-imported", "flags": { "frozen": true, "zone1": true }, "note": "Published adapted SOTA", "metrics": { "test": { "wer": 23.67, "csWer": 19.5 } } }, { "slug": "phowhisper-small-lora", "name": "PhoWhisper-small + LoRA", "org": "ViMedCSS paper", "paramsB": 0.244, "license": "BSD-3-Clause", "category": "vn-open", "track": "C", "provenance": "paper-imported", "flags": { "frozen": true, "zone1": true }, "metrics": { "test": { "wer": 27.13, "csWer": 30.26 } } }, { "slug": "vietasr", "name": "VietASR (Zipformer 68M)", "org": "SJTU / zzasdf", "paramsB": 0.068, "license": "Apache-2.0", "category": "vn-open", "track": "A", "provenance": "paper-imported", "flags": { "zone1": true }, "note": "Best published zero-shot", "metrics": { "test": { "wer": 27.56, "csWer": 58.38 } } }, { "slug": "phowhisper-large", "name": "PhoWhisper-large", "org": "VinAI", "paramsB": 1.55, "license": "BSD-3-Clause", "category": "vn-open", "track": "A", "provenance": "paper-imported", "flags": { "frozen": true, "zone1": true }, "note": "Clinical Scribe incumbent candidate", "metrics": { "test": { "wer": 31.24, "csWer": 55.05 } } }, { "slug": "whisper-large-v3", "name": "Whisper-large-v3", "org": "OpenAI", "paramsB": 1.55, "license": "Apache-2.0", "category": "multilingual-open", "track": "A", "provenance": "paper-imported", "flags": { "zone1": true }, "note": "Best published zero-shot CS-WER", "metrics": { "test": { "wer": 34.47, "csWer": 46.69 } } }, { "slug": "phowhisper-small", "name": "PhoWhisper-small", "org": "VinAI", "paramsB": 0.244, "license": "BSD-3-Clause", "category": "vn-open", "track": "A", "provenance": "paper-imported", "flags": { "frozen": true, "zone1": true }, "metrics": { "test": { "wer": 36.31 } } }, { "slug": "wav2vec2-base-vi", "name": "wav2vec2-base-vi", "org": "—", "paramsB": 0.095, "license": "Open", "category": "vn-open", "track": "A", "provenance": "paper-imported", "metrics": { "test": { "wer": 44.74 } } }, { "slug": "whisper-small", "name": "Whisper-small", "org": "OpenAI", "paramsB": 0.244, "license": "Apache-2.0", "category": "multilingual-open", "track": "A", "provenance": "paper-imported", "metrics": { "test": { "wer": 50.03 } } }, { "slug": "mms-1b", "name": "MMS-1B", "org": "Meta", "paramsB": 1, "license": "CC-BY-NC-4.0", "category": "multilingual-open", "track": "A", "provenance": "paper-imported", "metrics": { "test": { "wer": 58.3 } } }, { "slug": "zipformer-30m", "name": "Zipformer-30M RNNT (VLSP 2025)", "org": "hynt", "paramsB": 0.03, "license": "CC-BY-NC-ND-4.0", "category": "vn-open", "track": "A", "provenance": "platform-verified", "flags": { "referenceOnly": true }, "note": "VLSP 2025 challenge winner, run as the platform's reference yardstick", "metrics": { "test": { "wer": 19.81, "werCi": [ 19.06, 20.53 ], "cer": 15.92, "csWer": 74.55, "csWerCi": [ 72.35, 76.89 ], "nWer": 17.78, "mtrExact": 27.31, "mtrFuzzy": 36.48, "rtfx": 48.12 }, "hard": { "wer": 25.31, "werCi": [ 23.79, 26.9 ], "cer": 20.01, "csWer": 80.86, "csWerCi": [ 77.48, 84.09 ], "nWer": 23.02, "mtrExact": 19.81, "mtrFuzzy": 27.7, "rtfx": 54.72 } } }, { "slug": "qwen3-asr-0.6b", "name": "Qwen3-ASR-0.6B", "org": "Alibaba", "paramsB": 0.6, "license": "Apache-2.0", "category": "multilingual-open", "track": "A", "provenance": "platform-verified", "flags": { "zone1": true }, "note": "First sub-1B audio-LLM on the board — strongest code-switching recall of the verified runs", "metrics": { "test": { "wer": 21.29, "werCi": [ 20.5, 22.02 ], "cer": 16.11, "csWer": 68.49, "csWerCi": [ 66.12, 71.03 ], "nWer": 19.54, "mtrExact": 32.29, "mtrFuzzy": 42.31, "rtfx": 4.96 }, "hard": { "wer": 25, "werCi": [ 23.47, 26.4 ], "cer": 18.63, "csWer": 68.3, "csWerCi": [ 64.51, 72.18 ], "nWer": 23.22, "mtrExact": 30.76, "mtrFuzzy": 38, "rtfx": 8.82 } } }, { "slug": "gemini-2.5-pro", "name": "Gemini 2.5 Pro (audio)", "org": "Google", "license": "API", "category": "api-intl", "track": "A", "provenance": "planned", "flags": { "contamination": true }, "note": "Seeded this dataset labels — permanently contamination-flagged" }, { "slug": "gemini-3.5-flash", "name": "Gemini 3.5 Flash (audio)", "org": "Google", "license": "API", "category": "api-intl", "track": "A", "provenance": "api-snapshot", "flags": { "contamination": true }, "note": "Gemini-family — ViMedCSS labels were Gemini-2.5-seeded, so every Gemini result carries a permanent contamination flag", "metrics": { "test": { "wer": 13.84, "werCi": [ 13.06, 14.72 ], "cer": 12.22, "csWer": 29.38, "csWerCi": [ 26.95, 31.89 ], "nWer": 13.26, "mtrExact": 71.25, "mtrFuzzy": 80.75 }, "hard": { "wer": 16.08, "werCi": [ 14.46, 17.85 ], "cer": 14.12, "csWer": 24.66, "csWerCi": [ 21.25, 28.24 ], "nWer": 15.72, "mtrExact": 75.68, "mtrFuzzy": 80.03 } } }, { "slug": "deepgram-nova-3", "name": "Deepgram Nova-3", "org": "Deepgram", "license": "API", "category": "api-intl", "track": "A", "provenance": "api-snapshot", "note": "Latest Deepgram model with Vietnamese support; date-stamped API snapshot", "metrics": { "test": { "wer": 31.53, "werCi": [ 30, 33.14 ], "cer": 28.45, "csWer": 70.75, "csWerCi": [ 68.34, 73.47 ], "nWer": 30.07, "mtrExact": 29.99, "mtrFuzzy": 38.83 }, "hard": { "wer": 37.68, "werCi": [ 35.04, 40.26 ], "cer": 34.03, "csWer": 76.88, "csWerCi": [ 73.07, 80.31 ], "nWer": 36.06, "mtrExact": 23.67, "mtrFuzzy": 33.17 } } }, { "slug": "gpt-4o-transcribe", "name": "gpt-4o-transcribe", "org": "OpenAI", "license": "API", "category": "api-intl", "track": "A", "provenance": "planned" }, { "slug": "elevenlabs-scribe-v2", "name": "ElevenLabs Scribe v2", "org": "ElevenLabs", "license": "API", "category": "api-intl", "track": "B", "provenance": "planned", "note": "Keyterm biasing — Track B" }, { "slug": "fpt-ai-stt", "name": "FPT.AI STT", "org": "FPT Smart Cloud", "license": "API", "category": "api-vn", "track": "A", "provenance": "planned" } ] }, { "info": { "id": "vietmed", "name": "VietMed", "status": "coming-soon", "tagline": "Real-world Vietnamese medical speech recorded in actual doctor–patient conversations.", "source": "Real-world medical speech recordings" }, "models": [] }, { "info": { "id": "private-clinic", "name": "Private Clinic Set", "status": "coming-soon", "tagline": "Private conversational recordings from partner clinics, held out from public release to resist test-set contamination.", "source": "Partner clinic consultations", "eta": "2026" }, "models": [] } ] }