voice_audio_multi/model_specs/vibevoice_asr.json

{
  "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"
  ],
  "options": {
    "request": [
      {
        "name": "context",
        "type": "string",
        "description": "Context hint; context_info is also accepted.",
        "required": false
      },
      {
        "name": "language",
        "type": "string",
        "description": "ASR language label.",
        "required": false
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated text tokens.",
        "required": false,
        "min": 1,
        "default": 32768
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Sampling temperature; zero uses deterministic decoding.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "Nucleus probability in (0, 1].",
        "required": false,
        "default": 1
      },
      {
        "name": "top_k",
        "type": "int",
        "description": "Top-k sampling limit; zero disables filtering.",
        "required": false,
        "min": 0,
        "default": 50
      },
      {
        "name": "num_beams",
        "type": "int",
        "description": "Deterministic beam search width.",
        "required": false,
        "min": 1,
        "default": 1
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "description": "Positive generation repetition penalty.",
        "required": false,
        "default": 1
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Acoustic latent sampling seed.",
        "required": false,
        "min": 0,
        "default": 1234
      },
      {
        "name": "audio_chunk_mode",
        "type": "enum",
        "description": "Pre-encoder audio splitting; auto uses fixed chunks.",
        "required": false,
        "values": [
          "auto",
          "fixed",
          "vad",
          "none"
        ],
        "default": "auto"
      },
      {
        "name": "audio_chunk_duration_sec",
        "type": "float",
        "description": "Positive audio chunk duration in seconds.",
        "required": false,
        "default": 1200
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Weight storage type; component-specific settings inherit weight_type.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "tokenizer_weight_type",
        "type": "enum",
        "description": "Weight storage type; component-specific settings inherit weight_type.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "connector_weight_type",
        "type": "enum",
        "description": "Weight storage type; component-specific settings inherit weight_type.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "decoder_weight_type",
        "type": "enum",
        "description": "Weight storage type; component-specific settings inherit weight_type.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "tokenizer_weight_context_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "connector_weight_context_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 128
      },
      {
        "name": "decoder_weight_context_mb",
        "type": "int",
        "description": "Decoder weight context in MiB; 4096 on 64-bit builds, 1024 on 32-bit builds.",
        "required": false,
        "min": 1
      },
      {
        "name": "max_history_steps",
        "type": "int",
        "description": "Maximum retained decoder history; zero uses the model limit. Cannot exceed checkpoint position capacity; zero uses that capacity.",
        "required": false,
        "default": 0,
        "min": 0
      },
      {
        "name": "vad_model_path",
        "type": "path",
        "description": "VAD model (any built-in VAD family; default Silero VAD) used with audio_chunk_mode=vad.",
        "required": false,
        "default": "assets/framework/models/silero_vad"
      }
    ],
    "load": []
  },
  "capabilities": {
    "asr": [
      "segments",
      "speaker_turns",
      "vad_chunking"
    ]
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "ui": {
    "recommended_package": "vibevoice_asr_q8_0",
    "tags": [
      "ASR",
      "GGUF"
    ],
    "docs": [
      "docs/asr.md",
      "docs/gguf.md"
    ]
  },
  "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"
    }
  ],
  "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"
      }
    }
  ]
}