{
  "family": "chatterbox",
  "display_name": "Chatterbox",
  "description": "Open-source Chatterbox family for expressive TTS and voice conversion, with emotion exaggeration control, fast generation, zero-shot voice cloning, and an integrated multilingual TTS path.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts",
    "clone",
    "vc"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "ar",
    "da",
    "de",
    "el",
    "en",
    "es",
    "fi",
    "fr",
    "hi",
    "it",
    "ko",
    "ms",
    "nl",
    "no",
    "pl",
    "pt",
    "sv",
    "sw",
    "tr"
  ],
  "capabilities": {
    "clone": [
      "speaker_reference"
    ],
    "vc": [
      "speaker_reference"
    ]
  },
  "options": {
    "request": [
      {
        "name": "source_audio",
        "type": "audio_path",
        "description": "Source speech path for voice conversion; required for VC when no request audio input is supplied.",
        "required": false
      },
      {
        "name": "target_voice",
        "type": "audio_path",
        "description": "Target speaker reference path for voice conversion; required for VC when no request speaker audio is supplied.",
        "required": false
      },
      {
        "name": "exaggeration",
        "type": "float",
        "description": "Emotion and style exaggeration strength; default 0.5. Higher values produce more expressive speech and can speed up delivery.",
        "required": false,
        "min": 0.0,
        "default": 0.5
      },
      {
        "name": "guidance_scale",
        "type": "float",
        "description": "T3 classifier-free guidance scale; default 0.5. Lower values can improve pacing for fast reference speakers.",
        "required": false,
        "min": 0.0,
        "default": 0.5
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "T3 speech-token sampling temperature; default 0.8.",
        "required": false,
        "min": 0.0,
        "default": 0.8
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "description": "T3 repetition penalty applied during speech-token sampling; default 1.2.",
        "required": false,
        "min": 0.0,
        "default": 1.2
      },
      {
        "name": "min_p",
        "type": "float",
        "description": "T3 min-p sampling cutoff; default 0.05.",
        "required": false,
        "min": 0.0,
        "default": 0.05
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "T3 nucleus sampling probability; default 1.0 disables top-p truncation.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 1.0
      },
      {
        "name": "s3gen_guidance_scale",
        "type": "float",
        "description": "S3Gen token-to-mel flow guidance scale; default 0.7 for TTS and VC.",
        "required": false,
        "min": 0.0,
        "default": 0.7
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum T3 speech tokens for TTS/clone generation; default 384.",
        "required": false,
        "min": 1,
        "default": 384
      },
      {
        "name": "do_sample",
        "type": "bool",
        "description": "Enable stochastic T3 speech-token sampling; default true.",
        "required": false,
        "default": true
      },
      {
        "name": "stop_on_eos",
        "type": "bool",
        "description": "Stop T3 speech-token generation on EOS; default true.",
        "required": false,
        "default": true
      },
      {
        "name": "num_inference_steps",
        "type": "int",
        "description": "S3Gen flow inference steps for voice conversion; default 10.",
        "required": false,
        "min": 1,
        "default": 10
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Generation seed for reproducible T3, S3Gen, and vocoder sampling; omitted uses a random seed.",
        "required": false,
        "min": 0
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum characters per generated text chunk; default 128.",
        "required": false,
        "min": 1,
        "default": 128
      },
      {
        "name": "text_chunk_mode",
        "type": "enum",
        "description": "Framework text chunking mode for long-form TTS; default word_budget.",
        "preset": "text_chunk_mode_full",
        "required": false,
        "default": "word_budget"
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared Chatterbox component weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "t3_weight_type",
        "type": "enum",
        "description": "T3 weight storage type; defaults to weight_type.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "multilingual_t3",
        "type": "enum",
        "description": "Multilingual T3 checkpoint used for non-English TTS; default v2.",
        "values": [
          "v2",
          "v3"
        ],
        "required": false,
        "default": "v2"
      },
      {
        "name": "conditionals_cache_slots",
        "type": "int",
        "description": "Prepared voice-condition cache slots; default 1, set 0 to disable.",
        "required": false,
        "min": 0,
        "default": 1
      },
      {
        "name": "mem_saver",
        "type": "bool",
        "description": "Free non-conditional runtime graphs after each request chunk; default false.",
        "required": false,
        "default": false
      },
      {
        "name": "encoder_condition_samples",
        "type": "int",
        "description": "Reference audio samples used by the voice encoder at 16 kHz; default 96000, equal to 6 seconds.",
        "required": false,
        "min": 1,
        "default": 96000
      },
      {
        "name": "decoder_condition_samples",
        "type": "int",
        "description": "Reference audio samples used by the S3Gen decoder at 24 kHz; default 240000, equal to 10 seconds.",
        "required": false,
        "min": 1,
        "default": 240000
      },
      {
        "name": "t3_speech_cond_prompt_len",
        "type": "int",
        "description": "Maximum T3 speech-token prompt length kept from reference conditioning; default 150.",
        "required": false,
        "min": 0,
        "default": 150
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "english_tokenizer": "model:tokenizer.json",
        "multilingual_tokenizer": "model:grapheme_mtl_merged_expanded_v1.json",
        "cangjie_mapping": "model:Cangjie5_TC.json",
        "builtin_conditionals": "model:conds.pt"
      },
      "tensors": {
        "voice_encoder_weights": {
          "source": "weights:",
          "prefix": "voice_encoder"
        },
        "s3gen_weights": {
          "source": "weights:",
          "prefix": "s3gen"
        },
        "t3_english_weights": {
          "source": "weights:",
          "prefix": "t3_english"
        },
        "t3_multilingual_v2_weights": {
          "source": "weights:",
          "prefix": "t3_multilingual_v2"
        },
        "t3_multilingual_v3_weights": {
          "source": "weights:",
          "prefix": "t3_multilingual_v3"
        }
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "english_tokenizer": "model:tokenizer.json",
        "multilingual_tokenizer": "model:grapheme_mtl_merged_expanded_v1.json",
        "cangjie_mapping": "model:Cangjie5_TC.json",
        "builtin_conditionals": "model:conds.pt"
      },
      "tensors": {
        "voice_encoder_weights": "model:ve.safetensors",
        "s3gen_weights": "model:s3gen.safetensors",
        "t3_english_weights": "model:t3_cfg.safetensors",
        "t3_multilingual_v2_weights": "model:t3_mtl23ls_v2.safetensors",
        "t3_multilingual_v3_weights": "model:t3_mtl23ls_v3.safetensors"
      }
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "chatterbox_q8_0",
      "display_name": "Chatterbox Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Chatterbox-GGUF",
      "files": [
        "Chatterbox-GGUF/chatterbox-q8_0.gguf"
      ],
      "strip_prefix": "Chatterbox-GGUF"
    },
    {
      "id": "chatterbox_f16",
      "display_name": "Chatterbox F16 GGUF",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "Chatterbox-GGUF",
      "files": [
        "Chatterbox-GGUF/chatterbox-f16.gguf"
      ],
      "strip_prefix": "Chatterbox-GGUF"
    },
    {
      "id": "chatterbox_safetensors",
      "display_name": "Chatterbox Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "chatterbox",
      "files": [
        "ve.safetensors",
        "t3_cfg.safetensors",
        "s3gen.safetensors",
        "tokenizer.json"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "ResembleAI/chatterbox"
      }
    }
  ]
}