{
  "schema_version": 1,
  "family": "kokoro_tts",
  "display_name": "Kokoro 82M",
  "description": "Kokoro 82M multilingual text-to-speech with 54 preset voices, optimized native CPU inference, and shared eSpeak-ng phonemization.",
  "category": "tts",
  "status": "community",
  "tasks": ["tts"],
  "modes": ["offline"],
  "languages": ["en-us", "en-gb", "es", "fr-fr", "hi", "it", "ja", "pt-br", "zh"],
  "runtime": {
    "tags": ["gguf"]
  },
  "capabilities": {
    "tts": ["built_in_voices", "long_form"]
  },
  "options": {
    "request": [
      {
        "name": "language",
        "type": "string",
        "description": "Text language code; defaults to the selected voice prefix.",
        "required": false
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Decoder noise seed; omitted requests choose a random seed.",
        "required": false,
        "min": 0
      },
      {
        "name": "speed",
        "type": "float",
        "description": "Speech speed multiplier; default 1.0.",
        "required": false,
        "min": 0.01,
        "default": 1.0
      },
      {
        "name": "phonemes",
        "type": "string_list",
        "description": "Kokoro-vocabulary phonemes to synthesize directly, bypassing the built-in eSpeak-ng G2P. One entry per chunk: entries are rendered in order and merged into one result, so the caller chooses the split points that only its own G2P knows. Each entry must be at most 510 symbols and must not be empty -- an entry is a chunk to speak, so drop it rather than leaving it blank. Text is still required and its language must still match the voice.",
        "required": false
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum codepoints per text chunk before synthesis. Default 240.",
        "required": false,
        "min": 32,
        "default": 240
      },
      {
        "name": "return_timestamps",
        "type": "bool",
        "description": "Report where each phoneme group lands in the output, from the durations the model predicts before vocoding. Entries are phoneme groups in Kokoro's own alphabet, not written words, and do not map one-to-one onto the input text: see docs/models/kokoro_tts.md. Default false.",
        "required": false,
        "default": false
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "preset": "weight_type_full",
        "required": false,
        "default": "native",
        "description": "Storage type for matrix multiplication weights."
      },
      {
        "name": "conv_weight_type",
        "type": "enum",
        "preset": "weight_type_conv",
        "required": false,
        "default": "native",
        "description": "Storage type for convolution weights."
      },
      {
        "name": "weight_context_mb",
        "type": "int",
        "description": "Weight loading context size in MiB; default 512.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "predictor_duration_graph_mb",
        "type": "int",
        "description": "Predictor duration graph arena size in MiB; default 384.",
        "required": false,
        "min": 1,
        "default": 384
      },
      {
        "name": "predictor_text_graph_mb",
        "type": "int",
        "description": "Predictor text graph arena size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "predictor_tail_graph_mb",
        "type": "int",
        "description": "Predictor tail graph arena size in MiB; default 640.",
        "required": false,
        "min": 1,
        "default": 640
      },
      {
        "name": "graph_capacity_mode",
        "type": "enum",
        "description": "Offline graph capacity policy; default fixed.",
        "values": ["fixed", "tiered", "grow", "double"],
        "required": false,
        "default": "fixed"
      },
      {
        "name": "max_input_tokens",
        "type": "int",
        "description": "Predictor graph input token capacity; default min(512, model context length).",
        "required": false,
        "min": 1
      },
      {
        "name": "pre_tail_tokens",
        "type": "int",
        "description": "Optional prebuilt predictor tail token capacity; default 0.",
        "required": false,
        "min": 0,
        "default": 0
      }
    ],
    "load": []
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "kokoro_82m_q8_0",
      "display_name": "Kokoro 82M Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Kokoro-82M-GGUF",
      "files": [
        "Kokoro-82M-GGUF/kokoro-82m-q8_0.gguf"
      ],
      "strip_prefix": "Kokoro-82M-GGUF"
    },
    {
      "id": "kokoro_82m_bf16",
      "display_name": "Kokoro 82M BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "Kokoro-82M-GGUF",
      "files": [
        "Kokoro-82M-GGUF/kokoro-82m-bf16.gguf"
      ],
      "strip_prefix": "Kokoro-82M-GGUF"
    }
  ],
  "dependencies": [],
  "ui": {
    "recommended_package": "kokoro_82m_q8_0",
    "tags": ["TTS", "GGUF"],
    "builtin_voices": [
      "af_alloy", "af_aoede", "af_bella", "af_heart", "af_jessica", "af_kore", "af_nicole", "af_nova", "af_river", "af_sarah", "af_sky",
      "am_adam", "am_echo", "am_eric", "am_fenrir", "am_liam", "am_michael", "am_onyx", "am_puck", "am_santa",
      "bf_alice", "bf_emma", "bf_isabella", "bf_lily", "bm_daniel", "bm_fable", "bm_george", "bm_lewis",
      "ef_dora", "em_alex", "em_santa", "ff_siwis", "hf_alpha", "hf_beta", "hm_omega", "hm_psi", "if_sara", "im_nicola",
      "jf_alpha", "jf_gongitsune", "jf_nezumi", "jf_tebukuro", "jm_kumo", "pf_dora", "pm_alex", "pm_santa",
      "zf_xiaobei", "zf_xiaoni", "zf_xiaoxiao", "zf_xiaoyi", "zm_yunjian", "zm_yunxi", "zm_yunxia", "zm_yunyang"
    ],
    "default_voice": "af_heart",
    "docs": ["tests/kokoro_tts/MULTILINGUAL_GGUF.md"]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "voices": "model:voices.json",
        "vocabulary": "model:vocab.tsv"
      },
      "tensors": {
        "weights": {
          "source": "weights:",
          "prefix": "kokoro"
        }
      }
    }
  ]
}