voice_audio_multi/model_specs/maya1.json

{
  "schema_version": 1,
  "family": "maya1",
  "display_name": "Maya1",
  "description": "Maya Research's 3B expressive English text-to-speech model with natural-language voice design and inline emotion control.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts",
    "design"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "en"
  ],
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "capabilities": {
    "tts": [
      "voice_design",
      "emotion_control",
      "long_form"
    ],
    "design": [
      "voice_design"
    ]
  },
  "options": {
    "request": [
      {
        "name": "instruct",
        "type": "string",
        "description": "Natural-language description of the voice, accent, timbre, age, and delivery.",
        "required": true
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated audio-token count per text chunk; default 2048.",
        "required": false,
        "min": 1,
        "default": 2048
      },
      {
        "name": "min_tokens",
        "type": "int",
        "description": "Minimum generated audio-token count before EOS may stop generation; default 28.",
        "required": false,
        "min": 0,
        "default": 28
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Autoregressive sampling temperature; default 0.4. Must be > 0.",
        "required": false,
        "min": 0.0,
        "default": 0.4
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "Autoregressive nucleus sampling probability; must be in (0, 1], default 0.9.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 0.9
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "description": "Autoregressive repetition penalty; default 1.1. Must be > 0.",
        "required": false,
        "min": 0.0,
        "default": 1.1
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Sampling and SNAC decoder noise seed; omitted requests choose a random seed.",
        "required": false,
        "min": 0
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum Unicode codepoints per long-form text chunk; default 300.",
        "required": false,
        "min": 1,
        "default": 300
      },
      {
        "name": "text_chunk_mode",
        "type": "enum",
        "description": "Framework long-form text chunking mode; tag_aware preserves inline emotion tags.",
        "values": [
          "default",
          "tag_aware",
          "japanese",
          "endline"
        ],
        "required": false,
        "default": "tag_aware"
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared AR and SNAC decoder weight storage type; default native.",
        "values": [
          "native",
          "f32",
          "f16",
          "bf16",
          "q4_0",
          "q4_1",
          "q5_0",
          "q5_1",
          "q2_k",
          "q3_k",
          "q4_k",
          "q5_k",
          "q6_k",
          "q8_0",
          "nvfp4"
        ],
        "required": false,
        "default": "native"
      },
      {
        "name": "ar_weight_type",
        "type": "enum",
        "description": "AR backbone weight storage type; defaults to weight_type.",
        "values": [
          "native",
          "f32",
          "f16",
          "bf16",
          "q4_0",
          "q4_1",
          "q5_0",
          "q5_1",
          "q2_k",
          "q3_k",
          "q4_k",
          "q5_k",
          "q6_k",
          "q8_0",
          "nvfp4"
        ],
        "required": false
      },
      {
        "name": "codec_weight_type",
        "type": "enum",
        "description": "SNAC decoder weight storage type; only native or f32 is supported. Inherits weight_type when omitted; set codec_weight_type=f32 explicitly when using another shared weight_type.",
        "values": [
          "native",
          "f32"
        ],
        "required": false
      }
    ],
    "load": []
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "maya1_orig",
      "display_name": "Maya1 Original-Precision GGUF",
      "default": true,
      "format": "gguf",
      "precision": "orig",
      "target_directory": "Maya1-GGUF",
      "files": [
        "Maya1-GGUF/maya1-orig.gguf"
      ],
      "strip_prefix": "Maya1-GGUF"
    },
    {
      "id": "maya1_q8_0",
      "display_name": "Maya1 Q8_0 GGUF",
      "default": false,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Maya1-GGUF",
      "files": [
        "Maya1-GGUF/maya1-q8_0.gguf"
      ],
      "strip_prefix": "Maya1-GGUF"
    }
  ],
  "dependencies": [],
  "ui": {
    "recommended_package": "maya1_orig",
    "tags": [
      "TTS",
      "Design",
      "GGUF"
    ],
    "docs": [
      "docs/models/maya1.md",
      "docs/gguf.md"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_config": "model:tokenizer_config.json"
      },
      "tensors": {
        "model_weights": "weights:",
        "codec_weights": "weights:"
      }
    }
  ]
}