voice_audio_multi/model_specs/moss_tts_v15.json

{
  "schema_version": 1,
  "family": "moss_tts_v15",
  "display_name": "MOSS-TTS-v1.5",
  "description": "8B delay-pattern MOSS-TTS: zero-shot voice cloning from a short reference, 24 kHz, with an explicit duration budget.",
  "category": "community",
  "status": "experimental",
  "tasks": [
    "tts",
    "clone"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "en",
    "zh"
  ],
  "capabilities": {
    "tts": [
      "speaker_reference",
      "style_control",
      "long_form"
    ],
    "clone": [
      "speaker_reference"
    ]
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_merges": "model:merges.txt",
        "audio_tokenizer_config": "model:audio_tokenizer/config.json"
      },
      "tensors": {
        "model_weights": {
          "source": "weights:",
          "prefix": "model_weights"
        },
        "audio_tokenizer_weights": {
          "source": "weights:",
          "prefix": "audio_tokenizer_weights"
        }
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": ".",
        "audio_tokenizer": "audio_tokenizer"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_merges": "model:merges.txt",
        "audio_tokenizer_config": "audio_tokenizer:config.json"
      },
      "tensors": {
        "model_weights": "model:model.safetensors.index.json",
        "audio_tokenizer_weights": "audio_tokenizer:model.safetensors.index.json"
      }
    }
  ],
  "options": {
    "request": [
      {
        "name": "min_frames",
        "type": "int",
        "required": false,
        "default": 0,
        "description": "Minimum generated audio frames per text chunk; nonpositive values derive a text-length-based floor. Separate from duration_frames conditioning."
      },
      {
        "name": "max_frames",
        "type": "int",
        "required": false,
        "default": 0,
        "description": "Generated audio-frame ceiling per text chunk; nonpositive values derive a text-length-based ceiling. Runtime raises it to at least the effective min_frames plus the number of codebooks."
      },
      {
        "name": "instruct",
        "type": "string",
        "required": false,
        "description": "Voice description. Followed only loosely on this checkpoint; a reference recording is far more reliable."
      },
      {
        "name": "duration_frames",
        "type": "int",
        "required": false,
        "description": "Duration budget in codec frames at 12.5 a second, the model's own '- Tokens:' field. Nonpositive values disable duration conditioning."
      },
      {
        "name": "language",
        "type": "string",
        "required": false,
        "description": "Full language name; the model does not understand codes like 'en'."
      },
      {
        "name": "seed",
        "type": "int",
        "required": false,
        "description": "Reproduces a take exactly; default 0. Parsed as signed 32-bit integer then converted to uint32; negative values wrap.",
        "default": 0
      },
      {
        "name": "text_temperature",
        "type": "float",
        "required": false,
        "default": 1.5,
        "description": "Text-token sampling temperature. Must be finite and positive when sampling."
      },
      {
        "name": "temperature",
        "type": "float",
        "required": false,
        "description": "Audio sampling temperature; default 1.5. Must be finite and positive when sampling.",
        "default": 1.5
      },
      {
        "name": "top_p",
        "type": "float",
        "required": false,
        "description": "Audio nucleus sampling cutoff; default 0.6.",
        "default": 0.6
      },
      {
        "name": "top_k",
        "type": "int",
        "required": false,
        "description": "Audio top-k; default 50.",
        "default": 50
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "required": false,
        "description": "Audio repetition penalty; default 1.1. Must be finite and positive when history is nonempty.",
        "default": 1.1
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum Unicode codepoints per long-form text chunk; default 200.",
        "required": false,
        "min": 1,
        "default": 200
      },
      {
        "name": "text_chunk_mode",
        "type": "enum",
        "description": "Text chunking policy; word_budget also accepts default.",
        "values": [
          "word_budget",
          "tag_aware",
          "japanese",
          "endline"
        ],
        "required": false,
        "default": "word_budget"
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "values": [
          "native",
          "f32",
          "bf16",
          "q8_0"
        ],
        "required": false,
        "default": "native",
        "description": "Backbone and head weight storage; default native, which keeps a GGUF package's own type. Forcing bf16 dequantises a quantised package and costs the VRAM the quantisation was meant to save. f16 is rejected: this backbone produces NaN in it."
      }
    ],
    "load": []
  },
  "ui": {
    "tags": [
      "TTS",
      "Clone",
      "GGUF"
    ],
    "docs": [
      "docs/models/moss_tts_v15.md",
      "docs/tts.md",
      "docs/gguf.md"
    ],
    "recommended_package": "moss_tts_v15_q8_0_codec_f16"
  },
  "packages": [
    {
      "id": "moss_tts_v15_q8_0_codec_f16",
      "display_name": "MOSS-TTS-v1.5 Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "MOSS-TTS-v1.5-GGUF",
      "files": [
        "moss_tts_v15_q8_0_codec_f16.gguf"
      ]
    },
    {
      "id": "moss_tts_v15_q4_k_codec_f16",
      "display_name": "MOSS-TTS-v1.5 Q4_K GGUF",
      "default": false,
      "format": "gguf",
      "precision": "q4_k",
      "target_directory": "MOSS-TTS-v1.5-GGUF",
      "files": [
        "moss_tts_v15_q4_k_codec_f16.gguf"
      ]
    },
    {
      "id": "moss_tts_v15_bf16_codec_f16",
      "display_name": "MOSS-TTS-v1.5 BF16 GGUF",
      "default": false,
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "MOSS-TTS-v1.5-GGUF",
      "files": [
        "moss_tts_v15_bf16_codec_f16.gguf"
      ]
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "christopherthompson81/MOSS-TTS-v1.5-GGUF",
      "revision": "main",
      "gated": false
    }
  },
  "dependencies": []
}