{
  "schema_version": 1,
  "family": "magpie_tts",
  "display_name": "MagpieTTS Multilingual 357M",
  "description": "NVIDIA MagpieTTS is a multilingual neural text-to-speech model that generates NanoCodec audio tokens from text using baked speaker context embeddings, a decoder-only transformer, and an autoregressive local transformer before decoding the tokens to waveform audio.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "ar-AE",
    "ar-MSA",
    "ar-SA",
    "de",
    "en",
    "es",
    "fr",
    "hi",
    "it",
    "ko",
    "pt-BR",
    "vi",
    "zh"
  ],
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "capabilities": {
    "tts": [
      "long_form"
    ]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "Target synthesis language code used by the Magpie text frontend; default en.",
        "required": false,
        "default": "en"
      },
      {
        "name": "voice_id",
        "type": "enum",
        "description": "Packaged MagpieTTS voice name; default Aria. Numeric speaker index strings are also accepted within the checkpoint speaker count.",
        "values": [
          "Aria",
          "Jason",
          "John",
          "Leo",
          "Sofia"
        ],
        "required": false,
        "default": "Aria"
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Autoregressive audio-token sampling temperature; default 0.6 from the v2607 model config.",
        "required": false,
        "min": 0.0,
        "default": 0.6
      },
      {
        "name": "top_k",
        "type": "int",
        "description": "Autoregressive audio-token top-k sampling limit; default 80.",
        "required": false,
        "min": 1,
        "default": 80
      },
      {
        "name": "guidance_scale",
        "type": "float",
        "description": "Classifier-free guidance scale for Magpie decoder logits; default 2.5.",
        "required": false,
        "min": 0.0,
        "default": 2.5
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum Magpie decoder frames per text segment; default 500 from the v2607 model config.",
        "required": false,
        "min": 1,
        "default": 500
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum Unicode codepoints per long-form text chunk; default 100 for zh and ko, otherwise 300.",
        "required": false,
        "min": 1
      },
      {
        "name": "text_chunk_mode",
        "type": "enum",
        "description": "Framework text chunking mode.",
        "values": [
          "default",
          "tag_aware",
          "japanese",
          "endline"
        ],
        "required": false,
        "default": "default"
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Seed for audio-token sampling; default 0.",
        "required": false,
        "min": 0,
        "default": 0
      }
    ],
    "session": [
      {
        "name": "graph_arena_mb",
        "type": "int",
        "description": "Reusable ggml graph arena size in MiB for Magpie stages; default 1024.",
        "required": false,
        "min": 1,
        "default": 1024
      },
      {
        "name": "weight_context_mb",
        "type": "int",
        "description": "Weight loading context size in MiB; default 2048.",
        "required": false,
        "min": 1,
        "default": 2048
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Matmul weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "conv_weight_type",
        "type": "enum",
        "description": "Convolution weight storage type; default native.",
        "preset": "weight_type_conv",
        "required": false,
        "default": "native"
      }
    ],
    "load": []
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "magpie_tts_orig",
      "display_name": "MagpieTTS Multilingual 357M Original-Dtype GGUF",
      "default": false,
      "format": "gguf",
      "precision": "orig",
      "target_directory": "MagpieTTS-Multilingual-357M-GGUF",
      "files": [
        "MagpieTTS-Multilingual-357M-GGUF/magpie-tts-multilingual-357m-orig.gguf"
      ],
      "strip_prefix": "MagpieTTS-Multilingual-357M-GGUF"
    },
    {
      "id": "magpie_tts_q8_0",
      "display_name": "MagpieTTS Multilingual 357M Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "MagpieTTS-Multilingual-357M-GGUF",
      "files": [
        "MagpieTTS-Multilingual-357M-GGUF/magpie-tts-multilingual-357m-q8_0.gguf"
      ],
      "strip_prefix": "MagpieTTS-Multilingual-357M-GGUF"
    }
  ],
  "dependencies": [],
  "ui": {
    "recommended_package": "magpie_tts_q8_0",
    "tags": [
      "TTS",
      "GGUF"
    ],
    "docs": [
      "docs/models/magpie_tts.md",
      "docs/tts.md",
      "docs/gguf.md"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "codec_config": "model:codec_config.json",
        "speakers": "model:speakers.json"
      },
      "optional_files": {
        "ipa_dict": "model:ipa_dict.txt",
        "cmu_dict": "model:cmu_dict.txt",
        "german_dict": "model:german.dict",
        "german_heteronyms": "model:german.heteronym",
        "spanish_dict": "model:spanish.dict",
        "portuguese_dict": "model:portuguese.dict",
        "hindi_dict": "model:hindi.dict",
        "heteronyms": "model:heteronyms",
        "mandarin_chars": "model:mandarin_data/pinyin_chars.tsv",
        "mandarin_phrases": "model:mandarin_data/pinyin_phrases.tsv",
        "mandarin_jieba_dict": "model:mandarin_data/jieba.dict.utf8",
        "mandarin_hmm": "model:mandarin_data/hmm_model.utf8"
      },
      "tensors": {
        "model": {
          "source": "weights:",
          "prefix": "model"
        },
        "codec": {
          "source": "weights:",
          "prefix": "codec"
        }
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": ".",
        "codec": "../nemo_nano_codec_22khz_1_89kbps_safetensors"
      },
      "files": {
        "config": "model:config.json",
        "codec_config": "codec:config.json",
        "speakers": "model:41913ebaa70342058574da74293b7630_magpie_tts_multilingual_357m.nemo.speakers.json"
      },
      "optional_files": {
        "ipa_dict": "model:74b3e832fe914a569fcd42b51d06b2f1_ipa_dict_nv23.05.txt",
        "cmu_dict": "model:dc7d60d6b15a4651b21c9ca2932b62c6_ipa_cmudict-0.7b_nv23.01.txt",
        "german_dict": "model:cf01ab5c48c84f3282ef7888263361e5_de_nv230119.dict",
        "german_heteronyms": "model:c5e4ec2af5a14ce294f4b9edcc936535_de_nv230119.heteronym",
        "spanish_dict": "model:7dbc31751f224f2486090d59dc95b9f7_es_ES_nv230301.dict",
        "portuguese_dict": "model:05adc40366e149b69319acd4b28a4919_pt_br_prondict-v1.0.dict",
        "hindi_dict": "model:339da71c54b046f98cbcf38ef6d4ff67_hindi_phoneme_merged_phoneme_dict.dict",
        "heteronyms": "model:61d57a4ccb064d1e8e3d9f871c571932_heteronyms-052722"
      },
      "tensors": {
        "model": "model:model.safetensors",
        "codec": "codec:model.safetensors"
      }
    }
  ]
}