{
  "schema_version": 1,
  "family": "sense_asr",
  "display_name": "SenseVoice-Small",
  "description": "SenseVoice-Small multilingual speech recognition with rich event/emotion/ITN tags via a SAN-M encoder and CTC head, ported to audio.cpp.",
  "category": "asr",
  "status": "community",
  "tasks": [
    "asr"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "auto",
    "zh",
    "en",
    "yue",
    "ja",
    "ko",
    "pt",
    "ru",
    "es",
    "it",
    "fr",
    "de",
    "nl",
    "pl",
    "tr",
    "ar",
    "hi",
    "vi",
    "th",
    "id",
    "ms",
    "fa",
    "nospeech"
  ],
  "capabilities": {
    "asr": [
      "vad_chunking",
      "partial_results"
    ]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "Recognition language, or auto to let the model infer it from the audio.",
        "required": false,
        "default": "auto"
      },
      {
        "name": "enable_itn",
        "type": "bool",
        "description": "Enable inverse text normalization (adds the withitn query token).",
        "required": false,
        "default": true
      },
      {
        "name": "keep_tags",
        "type": "bool",
        "description": "Keep <|event|>/<|emotion|>/<|language|> tags inline in the output text.",
        "required": false,
        "default": false
      },
      {
        "name": "audio_chunk_mode",
        "type": "enum",
        "description": "Audio chunking mode: auto, fixed, or none.",
        "values": [
          "auto",
          "fixed",
          "none"
        ],
        "required": false,
        "default": "auto"
      },
      {
        "name": "audio_chunk_duration_sec",
        "type": "float",
        "description": "Fixed chunk duration in seconds when not using VAD segmentation.",
        "required": false,
        "min": 0.001,
        "default": 30
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared model weight storage type.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "encoder_graph_arena_mb",
        "type": "int",
        "description": "Encoder graph arena size in MB.",
        "required": false,
        "min": 64,
        "default": 1024
      },
      {
        "name": "vad_model_path",
        "type": "string",
        "description": "Path to the Silero VAD model directory used by automatic audio chunking.",
        "required": false,
        "default": "assets/framework/models/silero_vad"
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf",
      "server",
      "stream",
      "cuda",
      "metal",
      "cpu"
    ]
  },
  "packages": [
    {
      "id": "sensevoice_small_q8",
      "display_name": "SenseVoice-Small Q8 GGUF",
      "description": "audio.cpp GGUF built from the SenseVoice-Small checkpoint via the SenseVoice llama.cpp export script.",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "SenseVoice-Small-GGUF",
      "files": [
        "sensevoice-small-q8-audiocpp-v1.gguf"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "FunAudioLLM/SenseVoiceSmall-GGUF-audiocpp",
        "revision": "5c3fcfe748a8714216bc135476d5863084fddb72",
        "gated": false
      }
    }
  ],
  "dependencies": [],
  "ui": {
    "recommended_package": "sensevoice_small_q8",
    "tags": [
      "ASR",
      "GGUF",
      "Stream"
    ],
    "docs": [
      "docs/community_models/sense_asr.md"
    ],
    "summary": "SenseVoice-Small transcription with event/emotion tags and ITN."
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {},
      "optional_files": {},
      "tensors": {
        "weights": "weights:"
      }
    }
  ]
}