voice_audio_multi/model_specs/moss_ttsd.json

{
  "schema_version": 1,
  "family": "moss_ttsd",
  "display_name": "MOSS-TTSD",
  "description": "8B delay-pattern dialogue TTS: speaker-tagged conversation in one take, with per-speaker zero-shot voice cloning. 24 kHz, 16 codebooks.",
  "category": "community",
  "status": "experimental",
  "tasks": [
    "tts",
    "clone"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "en",
    "zh"
  ],
  "capabilities": {
    "tts": [
      "speaker_reference",
      "multi_speaker",
      "long_form"
    ],
    "clone": [
      "speaker_reference",
      "multi_speaker"
    ]
  },
  "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": "voice_samples",
        "type": "string",
        "required": false,
        "description": "Per-speaker reference WAV paths, comma separated and positional: the first is [S1], the second [S2]. Leave an entry empty to let that speaker be invented rather than cloned. Cannot be combined with voice_ref."
      },
      {
        "name": "reference_text",
        "type": "string",
        "required": false,
        "description": "What the reference recordings say, speaker-tagged. Prepended to the dialogue so the continuation carries on from words it has already spoken."
      },
      {
        "name": "language",
        "type": "string",
        "required": false,
        "description": "Full language name; the model does not understand codes like 'en'."
      },
      {
        "name": "instruct",
        "type": "string",
        "required": false,
        "description": "Free-text direction for the delivery."
      },
      {
        "name": "seed",
        "type": "int",
        "required": false,
        "description": "Reproduces a take exactly. Parsed as signed 32-bit integer then converted to uint32; negative values wrap."
      },
      {
        "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,
        "min": 0.0,
        "default": 1.5,
        "description": "Audio sampling temperature; must be > 0, default 1.5."
      },
      {
        "name": "top_p",
        "type": "float",
        "required": false,
        "min": 0.0,
        "default": 0.6,
        "description": "Audio nucleus sampling cutoff."
      },
      {
        "name": "top_k",
        "type": "int",
        "required": false,
        "default": 50,
        "description": "Audio top-k. Nonpositive values disable filtering."
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "required": false,
        "min": 0.0,
        "default": 1.1,
        "description": "Audio repetition penalty; must be > 0. Checked when token history is present; default 1.1."
      },
      {
        "name": "max_frames",
        "type": "int",
        "required": false,
        "description": "Ceiling on generated codec frames at 12.5/s; 0 derives one from the dialogue length. Nonpositive values derive an automatic ceiling."
      }
    ],
    "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_ttsd.md",
      "docs/tts.md",
      "docs/gguf.md"
    ],
    "recommended_package": "moss_ttsd_q8_0_codec_f16"
  },
  "packages": [
    {
      "id": "moss_ttsd_q8_0_codec_f16",
      "display_name": "MOSS-TTSD Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "MOSS-TTSD-GGUF",
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "christopherthompson81/MOSS-TTSD-GGUF"
      },
      "files": [
        "moss_ttsd_q8_0_codec_f16.gguf"
      ]
    },
    {
      "id": "moss_ttsd_q4_k_codec_f16",
      "display_name": "MOSS-TTSD Q4_K GGUF (f16 heads)",
      "default": false,
      "format": "gguf",
      "precision": "q4_k",
      "target_directory": "MOSS-TTSD-GGUF",
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "christopherthompson81/MOSS-TTSD-GGUF"
      },
      "files": [
        "moss_ttsd_q4_k_codec_f16.gguf"
      ]
    },
    {
      "id": "moss_ttsd_bf16_codec_f16",
      "display_name": "MOSS-TTSD BF16 GGUF",
      "default": false,
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "MOSS-TTSD-GGUF",
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "christopherthompson81/MOSS-TTSD-GGUF"
      },
      "files": [
        "moss_ttsd_bf16_codec_f16.gguf"
      ]
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "christopherthompson81/MOSS-TTS-v1.5-GGUF",
      "revision": "main",
      "gated": false
    }
  },
  "dependencies": []
}