voice_audio_multi/model_specs/firered_vad.json

{
  "schema_version": 1,
  "family": "firered_vad",
  "display_name": "FireRed VAD",
  "description": "DFSMN speech activity detection with offline and causal Stream-VAD checkpoints.",
  "category": "audio_tools",
  "status": "experimental",
  "tasks": [
    "vad"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "language_agnostic"
  ],
  "capabilities": {
    "vad": [
      "speech_segments"
    ]
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "options": {
    "request": [
      {
        "name": "threshold",
        "type": "float",
        "description": "Speech probability threshold; defaults to 0.4 for VAD and 0.5 for Stream-VAD.",
        "required": false,
        "min": 0,
        "max": 1
      },
      {
        "name": "smooth_window_ms",
        "type": "int",
        "description": "Trailing probability smoothing window in milliseconds, rounded to the nearest 10 ms. Validation occurs after conversion to 10-ms frames; metadata bounds describe intended settings.",
        "required": false,
        "default": 50,
        "min": 10
      },
      {
        "name": "min_speech_duration_ms",
        "type": "int",
        "description": "Speech confirmation duration, rounded to the nearest 10 ms; defaults to 200 ms for VAD and 80 ms for Stream-VAD. Validation occurs after conversion to 10-ms frames; metadata bounds describe intended settings.",
        "required": false,
        "min": 10
      },
      {
        "name": "max_speech_duration_sec",
        "type": "float",
        "description": "Maximum speech duration in seconds, rounded to the nearest 10 ms. Validation occurs after conversion to 10-ms frames; metadata bounds describe intended settings.",
        "required": false,
        "default": 20,
        "min": 0.02
      },
      {
        "name": "min_silence_duration_ms",
        "type": "int",
        "description": "Silence confirmation duration in milliseconds, rounded to the nearest 10 ms. Validation occurs after conversion to 10-ms frames; metadata bounds describe intended settings.",
        "required": false,
        "default": 200,
        "min": 0
      },
      {
        "name": "merge_silence_duration_ms",
        "type": "int",
        "description": "Merge shorter internal silence gaps in offline VAD; milliseconds rounded to the nearest 10 ms.",
        "required": false,
        "default": 0,
        "min": 0
      },
      {
        "name": "speech_pad_ms",
        "type": "int",
        "description": "Extend offline VAD segments on each side; milliseconds rounded to the nearest 10 ms.",
        "required": false,
        "default": 0,
        "min": 0
      },
      {
        "name": "speech_start_pad_ms",
        "type": "int",
        "description": "Stream-VAD start padding in milliseconds, rounded to the nearest 10 ms and at least the smoothing window.",
        "required": false,
        "default": 50,
        "min": 0
      },
      {
        "name": "audio_chunk_duration_sec",
        "type": "float",
        "description": "Maximum network duration per offline chunk in seconds, rounded to the nearest 10 ms; each chunk resets network history, matching upstream.",
        "required": false,
        "default": 300,
        "min": 0.01
      }
    ],
    "session": [],
    "load": []
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "firered_vad_f32",
      "display_name": "FireRed VAD GGUF F32",
      "default": true,
      "format": "gguf",
      "precision": "f32",
      "target_directory": "FireRedVAD-GGUF",
      "files": [
        "FireRedVAD-GGUF/firered-vad-f32.gguf"
      ],
      "strip_prefix": "FireRedVAD-GGUF"
    },
    {
      "id": "firered_stream_vad_f32",
      "display_name": "FireRed Stream-VAD GGUF F32",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "FireRedVAD-GGUF",
      "files": [
        "FireRedVAD-GGUF/firered-stream-vad-f32.gguf"
      ],
      "strip_prefix": "FireRedVAD-GGUF"
    }
  ],
  "dependencies": [],
  "ui": {
    "tags": [
      "VAD",
      "GGUF"
    ],
    "docs": [
      "docs/models/firered_vad.md"
    ],
    "recommended_package": "firered_vad_f32"
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "tensors": {
        "weights": {
          "source": "weights:",
          "prefix": "weights"
        }
      }
    }
  ]
}