voice_audio_multi/model_specs/granite5asr.json

{
  "family": "granite5asr",
  "display_name": "Granite Speech 5.0 470M TurboCTC",
  "description": "IBM Granite Speech 5.0 TurboCTC 470M parameter Conformer CTC model. The current checkpoint transcribes English, French, German, Spanish, Portuguese, Italian, and Japanese with punctuation and capitalization. An older English-only checkpoint still loads.",
  "category": "asr",
  "status": "supported",
  "tasks": [
    "asr"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "en",
    "fr",
    "de",
    "es",
    "pt",
    "it",
    "ja"
  ],
  "capabilities": {
    "asr": [
      "vad_chunking",
      "partial_results",
      "word_timestamps"
    ]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "Accepted and ignored. Granite has no language input. A result is labelled en unless decoding emits a trained language token such as <|fr|>.",
        "required": false,
        "default": "en"
      },
      {
        "name": "audio_chunk_mode",
        "type": "enum",
        "description": "Offline chunk mode: auto, fixed, vad, or none. VAD requires a regular model file; otherwise fixed chunks are used.",
        "values": [
          "auto",
          "fixed",
          "vad",
          "none"
        ],
        "required": false,
        "default": "auto"
      },
      {
        "name": "audio_chunk_duration_sec",
        "type": "float",
        "description": "Positive maximum audio chunk duration for fixed and VAD chunking.",
        "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. Positive MiB allocation; invalid/nonpositive values resolve to 1024 MiB.",
        "required": false,
        "min": 1,
        "default": 1024
      },
      {
        "name": "vad_model_path",
        "type": "string",
        "description": "Path to a VAD weight file of any built-in VAD family (e.g. silero_vad_16k.safetensors), not a directory; VAD runs only when this path is a regular file.",
        "required": false
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf",
      "server",
      "stream",
      "cuda",
      "metal",
      "cpu"
    ]
  },
  "ui": {
    "recommended_package": "granite5asr_q8_0",
    "tags": [
      "ASR",
      "GGUF",
      "Stream"
    ],
    "docs": [
      "docs/models/granite5asr.md",
      "docs/asr.md",
      "docs/gguf.md"
    ],
    "summary": "IBM Granite Speech 5.0 470M TurboCTC multilingual transcription."
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "ibm-granite/granite-speech-5.0-470m-turboctc",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "granite5asr_q8_0",
      "display_name": "Granite Speech 5.0 470M TurboCTC Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Granite-Speech-5.0-470M-TurboCTC-GGUF",
      "files": [
        "Granite-Speech-5.0-470M-TurboCTC-GGUF/granite-speech-5.0-470m-turboctc-q8_0.gguf"
      ],
      "strip_prefix": "Granite-Speech-5.0-470M-TurboCTC-GGUF",
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "audio-cpp/audio.cpp-gguf",
        "revision": "main",
        "gated": false
      }
    },
    {
      "id": "granite5asr_safetensors",
      "display_name": "Granite Speech 5.0 470M TurboCTC Native Safetensors",
      "format": "safetensors",
      "precision": "bfloat16",
      "target_directory": "granite5asr",
      "files": [
        "config.json",
        "tokenizer.json",
        "model.safetensors"
      ]
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json"
      },
      "optional_files": {
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "tensors": {
        "weights": "model:model.safetensors"
      }
    }
  ]
}