{
  "family": "vibevoice_asr",
  "display_name": "VibeVoice ASR",
  "description": "Microsoft long-form speech-to-text model that processes up to 60 minutes of audio in one pass and produces structured transcripts with speakers, timestamps, content, hotwords, and 50+ language support.",
  "category": "asr",
  "status": "supported",
  "tasks": [
    "asr"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "auto",
    "51 languages"
  ],
  "capabilities": {
    "asr": [
      "segments",
      "speaker_turns",
      "vad_chunking"
    ]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "Optional transcript language label; VibeVoice-ASR can auto-detect language and handle code-switching without an explicit language setting.",
        "required": false
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated text tokens; default 32768.",
        "required": false,
        "min": 1,
        "default": 32768
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Sampling temperature; default 0 uses deterministic decoding.",
        "required": false,
        "min": 0.0,
        "default": 0.0
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "Nucleus sampling probability; default 1.0.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 1.0
      },
      {
        "name": "top_k",
        "type": "int",
        "description": "Top-k sampling limit; default 50, 0 disables top-k filtering.",
        "required": false,
        "min": 0,
        "default": 50
      },
      {
        "name": "num_beams",
        "type": "int",
        "description": "Beam count for deterministic beam search; default 1.",
        "required": false,
        "min": 1,
        "default": 1
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "description": "Generation repetition penalty; must be positive, default 1.0.",
        "required": false,
        "default": 1.0
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Acoustic latent sampling seed for reproducible output; default 1234.",
        "required": false,
        "min": 0,
        "default": 1234
      },
      {
        "name": "audio_chunk_mode",
        "type": "enum",
        "description": "Audio chunking mode; default auto uses fixed chunks.",
        "values": [
          "auto",
          "fixed",
          "vad",
          "none"
        ],
        "required": false,
        "default": "auto"
      },
      {
        "name": "audio_chunk_duration_sec",
        "type": "float",
        "description": "Audio chunk duration in seconds for fixed and VAD chunking; must be positive, default 1200.",
        "required": false,
        "default": 1200.0
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Tokenizer, connector, and decoder weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "tokenizer_weight_type",
        "type": "enum",
        "description": "Speech tokenizer weight storage type; defaults to weight_type when set, otherwise native.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "connector_weight_type",
        "type": "enum",
        "description": "Acoustic and semantic connector weight storage type; defaults to weight_type when set, otherwise native.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "decoder_weight_type",
        "type": "enum",
        "description": "Text decoder weight storage type; defaults to weight_type when set, otherwise native.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "tokenizer_weight_context_mb",
        "type": "int",
        "description": "Speech tokenizer weight context arena size in MiB; default 512.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "connector_weight_context_mb",
        "type": "int",
        "description": "Acoustic and semantic connector weight context arena size in MiB; default 128.",
        "required": false,
        "min": 1,
        "default": 128
      },
      {
        "name": "decoder_weight_context_mb",
        "type": "int",
        "description": "Text decoder weight context arena size in MiB; default 4096.",
        "required": false,
        "min": 1,
        "default": 4096
      },
      {
        "name": "vad_path",
        "type": "path",
        "description": "Silero VAD model path used by audio_chunk_mode=vad.",
        "required": false
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_vocab": "model:vocab.json",
        "tokenizer_merges": "model:merges.txt"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "tensors": {
        "model_weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_vocab": "model:vocab.json",
        "tokenizer_merges": "model:merges.txt"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "tensors": {
        "model_weights": "model:model.safetensors.index.json"
      }
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "vibevoice_asr_q8_0",
      "display_name": "VibeVoice ASR Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "VibeVoice-ASR-GGUF",
      "files": [
        "VibeVoice-ASR-GGUF/vibevoice-asr-q8_0.gguf"
      ],
      "strip_prefix": "VibeVoice-ASR-GGUF"
    },
    {
      "id": "vibevoice_asr_f16",
      "display_name": "VibeVoice ASR F16 GGUF",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "VibeVoice-ASR-GGUF",
      "files": [
        "VibeVoice-ASR-GGUF/vibevoice-asr-f16.gguf"
      ],
      "strip_prefix": "VibeVoice-ASR-GGUF"
    }
  ],
  "dependencies": [
    {
      "kind": "bundled_model",
      "family": "silero_vad",
      "path": "assets/framework/models/silero_vad",
      "scope": "session",
      "option": "vad_path",
      "required": false,
      "required_when": [
        {
          "scope": "request",
          "option_key": "audio_chunk_mode",
          "equals": "vad"
        }
      ]
    }
  ]
}