voice_audio_multi/model_specs/vibevoice.json

{
  "family": "vibevoice",
  "display_name": "VibeVoice",
  "description": "Microsoft long-form multi-speaker TTS model for expressive conversational audio such as podcasts, supporting up to 90 minutes of speech with as many as four speakers.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "en",
    "zh"
  ],
  "options": {
    "request": [
      {
        "name": "voice_samples",
        "type": "string",
        "description": "Comma-separated reference WAV paths ordered by speaker ID; cannot be combined with voice_ref.",
        "required": false
      },
      {
        "name": "num_inference_steps",
        "type": "int",
        "description": "Diffusion steps; defaults to the checkpoint configuration.",
        "required": false,
        "min": 1
      },
      {
        "name": "guidance_scale",
        "type": "float",
        "description": "Classifier-free guidance scale.",
        "required": false,
        "min": 0,
        "default": 1.3
      },
      {
        "name": "max_length_times",
        "type": "float",
        "description": "Positive generation length multiplier.",
        "required": false,
        "default": 2
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated tokens; zero uses the model-derived limit.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "do_sample",
        "type": "bool",
        "description": "Enable sampling.",
        "required": false,
        "default": false
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Positive sampling temperature.",
        "required": false,
        "default": 1
      },
      {
        "name": "top_k",
        "type": "int",
        "description": "Top-k sampling limit; zero disables filtering.",
        "required": false,
        "min": 0,
        "default": 50
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "Nucleus probability in (0, 1].",
        "required": false,
        "default": 1
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Sampling seed.",
        "required": false,
        "min": 0,
        "max": 4294967295,
        "default": 1234
      },
      {
        "name": "prompt_noise_file",
        "type": "path",
        "description": "F32 prompt acoustic noise file.",
        "required": false
      },
      {
        "name": "diffusion_noise_file",
        "type": "path",
        "description": "F32 initial diffusion noise file.",
        "required": false
      }
    ],
    "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": "diffusion_head_weight_type",
        "type": "enum",
        "description": "Weight storage type; component-specific settings inherit weight_type.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "lora",
        "type": "path",
        "description": "Decoder adapter file or PEFT directory applied when creating the session; do not also supply the load option.",
        "required": false
      },
      {
        "name": "lora_scale",
        "type": "float",
        "description": "Session adapter scale override; omitted uses lora_alpha / r. Must be strictly positive when an adapter is supplied.",
        "required": false
      }
    ],
    "load": [
      {
        "name": "lora",
        "type": "path",
        "description": "Decoder adapter file or PEFT directory, merged at load time.",
        "required": false
      },
      {
        "name": "lora_scale",
        "type": "float",
        "description": "LoRA scale override; omitted uses lora_alpha / r. Must be strictly positive when an adapter is supplied.",
        "required": false
      }
    ]
  },
  "capabilities": {
    "tts": [
      "multi_speaker",
      "long_form"
    ]
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "ui": {
    "recommended_package": "vibevoice_1_5b_q8_0",
    "tags": [
      "TTS",
      "GGUF"
    ],
    "docs": [
      "docs/tts.md",
      "docs/gguf.md"
    ]
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "vibevoice_1_5b_q8_0",
      "display_name": "VibeVoice 1.5B Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "VibeVoice-1.5B-GGUF",
      "files": [
        "VibeVoice-1.5B-GGUF/vibevoice-1.5b-q8_0.gguf"
      ],
      "strip_prefix": "VibeVoice-1.5B-GGUF"
    },
    {
      "id": "vibevoice_1_5b_bf16",
      "display_name": "VibeVoice 1.5B BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "VibeVoice-1.5B-GGUF",
      "files": [
        "VibeVoice-1.5B-GGUF/vibevoice-1.5b-bf16.gguf"
      ],
      "strip_prefix": "VibeVoice-1.5B-GGUF"
    },
    {
      "id": "vibevoice_7b_q8_0",
      "display_name": "VibeVoice 7B Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "VibeVoice-7B-GGUF",
      "files": [
        "vibevoice-7b-q8_0.gguf"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "audio-cpp/VibeVoice-7B-GGUF",
        "revision": "main",
        "gated": false
      }
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "preprocessor_config": "model:preprocessor_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_vocab": "model:vocab.json",
        "tokenizer_merges": "model:merges.txt"
      },
      "tensors": {
        "model_weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "preprocessor_config": "model:preprocessor_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_vocab": "model:vocab.json",
        "tokenizer_merges": "model:merges.txt"
      },
      "tensors": {
        "model_weights": "model:model.safetensors.index.json"
      }
    }
  ]
}