{
  "family": "pocket_tts",
  "display_name": "PocketTTS",
  "description": "Kyutai 100M-parameter CPU-friendly TTS package set for real-time local synthesis and small-footprint voice cloning in English, German, Italian, Portuguese, and Spanish.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts",
    "clone"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "en",
    "de",
    "it",
    "pt",
    "es"
  ],
  "capabilities": {
    "clone": [
      "speaker_reference"
    ]
  },
  "options": {
    "request": [
      {
        "name": "max_steps",
        "type": "int",
        "description": "Maximum acoustic generation frames per text chunk; default 0 estimates the cap from text length.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Text-conditioning token budget accepted by the PocketTTS generation path; default 50.",
        "required": false,
        "min": 1,
        "default": 50
      },
      {
        "name": "frames_after_eos",
        "type": "int",
        "description": "Acoustic frames to continue after the EOS head fires; default -1 uses the model recommendation or the text-length heuristic.",
        "required": false,
        "min": -1,
        "default": -1
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Positive acoustic sampling temperature. When omitted, PocketTTS uses the model config default: 0.3 for English built-in defaults and 0.7 for other built-in defaults.",
        "required": false,
        "min": 0.0
      },
      {
        "name": "noise_clamp",
        "type": "float",
        "description": "Clamp sampled latent noise to this positive magnitude; default -1 disables clamping.",
        "required": false,
        "default": -1.0
      },
      {
        "name": "eos_threshold",
        "type": "float",
        "description": "EOS logit threshold used to stop acoustic generation; default -4.0.",
        "required": false,
        "default": -4.0
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Acoustic noise seed; omitted requests choose a random seed during session preparation.",
        "required": false,
        "min": 0
      },
      {
        "name": "noise_path",
        "type": "path",
        "description": "Optional binary float latent-noise schedule. The schedule must provide at least max_steps latent vectors.",
        "required": false
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum UTF-8 codepoints per long-form text chunk; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "voice_state_path",
        "type": "path",
        "description": "Optional prepared PocketTTS voice-state file to use instead of a named preset or reference audio.",
        "required": false
      },
      {
        "name": "truncate_clone_audio",
        "type": "bool",
        "description": "Limit reference audio used for voice cloning to the first 30 seconds; default false.",
        "required": false,
        "default": false
      }
    ],
    "session": [
      {
        "name": "voice_state_cache_slots",
        "type": "int",
        "description": "Prepared voice-state cache slots; default 4, set 0 to disable.",
        "required": false,
        "min": 0,
        "default": 4
      },
      {
        "name": "voice_state_export_path",
        "type": "path",
        "description": "Optional path where session preparation exports the prepared voice state for later reuse.",
        "required": false
      },
      {
        "name": "graph_capacity_mode",
        "type": "enum",
        "description": "Graph capacity allocation strategy; default tiered on CPU host-graph backends and double otherwise.",
        "values": [
          "fixed",
          "tiered",
          "grow",
          "double"
        ],
        "required": false
      },
      {
        "name": "prompt_graph_capacity",
        "type": "int",
        "description": "Fixed prompt graph capacity in acoustic prompt steps; default 0 lets the graph-capacity mode choose.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "generation_graph_capacity",
        "type": "int",
        "description": "Fixed acoustic generation graph capacity in generated frames; default 0 lets the graph-capacity mode choose.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "weight_context_mb",
        "type": "int",
        "description": "Default weight context size in MiB for FlowLM and Mimi weights; default 64.",
        "required": false,
        "min": 1,
        "default": 64
      },
      {
        "name": "flow_weight_context_mb",
        "type": "int",
        "description": "FlowLM weight context size in MiB; default inherits weight_context_mb.",
        "required": false,
        "min": 1
      },
      {
        "name": "mimi_encoder_weight_context_mb",
        "type": "int",
        "description": "Mimi encoder weight context size in MiB; default inherits weight_context_mb.",
        "required": false,
        "min": 1
      },
      {
        "name": "mimi_decoder_weight_context_mb",
        "type": "int",
        "description": "Mimi decoder weight context size in MiB; default inherits weight_context_mb.",
        "required": false,
        "min": 1
      },
      {
        "name": "flow_weight_view_context_mb",
        "type": "int",
        "description": "FlowLM temporary weight-view context size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "flow_step_graph_arena_mb",
        "type": "int",
        "description": "FlowLM step graph arena size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "mimi_encoder_graph_arena_mb",
        "type": "int",
        "description": "Mimi encoder graph arena size in MiB; default 512.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "mimi_conv_graph_arena_mb",
        "type": "int",
        "description": "Mimi decoder convolution graph arena size in MiB; default 32.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "mimi_transformer_graph_arena_mb",
        "type": "int",
        "description": "Mimi decoder transformer graph arena size in MiB; default 96.",
        "required": false,
        "min": 1,
        "default": 96
      },
      {
        "name": "mimi_tail_graph_arena_mb",
        "type": "int",
        "description": "Mimi decoder tail graph arena size in MiB; default 512.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "mimi_full_chunk_frames",
        "type": "int",
        "description": "Mimi full-sequence decoder chunk size in codec frames; default 90.",
        "required": false,
        "min": 1,
        "default": 90
      },
      {
        "name": "mimi_stage2_chunk_frames",
        "type": "int",
        "description": "Mimi stage-2 decoder chunk size in codec frames; default 900.",
        "required": false,
        "min": 1,
        "default": 900
      },
      {
        "name": "use_full_mimi",
        "type": "bool",
        "description": "Use the full-sequence Mimi decoder path instead of the streaming chunked path; default true.",
        "required": false,
        "default": true
      },
      {
        "name": "matmul_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": []
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "tokenizer": "model:tokenizer.model"
      },
      "optional_files": {
        "config": "model:config.yaml"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "language": "languages/english"
      },
      "files": {
        "tokenizer": "language:tokenizer.model"
      },
      "optional_files": {
        "config": "language:config.yaml"
      },
      "tensors": {
        "weights": "language:model.safetensors"
      }
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "pocket_tts_english_q8_0",
      "display_name": "PocketTTS English Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "PocketTTS-GGUF/english",
      "files": [
        "PocketTTS-GGUF/english/pocket-tts-english-q8_0.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/english"
    },
    {
      "id": "pocket_tts_english_bf16",
      "display_name": "PocketTTS English BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "PocketTTS-GGUF/english",
      "files": [
        "PocketTTS-GGUF/english/pocket-tts-english-bf16.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/english"
    },
    {
      "id": "pocket_tts_german_q8_0",
      "display_name": "PocketTTS German Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "PocketTTS-GGUF/german",
      "files": [
        "PocketTTS-GGUF/german/pocket-tts-german-q8_0.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/german"
    },
    {
      "id": "pocket_tts_german_bf16",
      "display_name": "PocketTTS German BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "PocketTTS-GGUF/german",
      "files": [
        "PocketTTS-GGUF/german/pocket-tts-german-bf16.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/german"
    },
    {
      "id": "pocket_tts_italian_q8_0",
      "display_name": "PocketTTS Italian Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "PocketTTS-GGUF/italian",
      "files": [
        "PocketTTS-GGUF/italian/pocket-tts-italian-q8_0.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/italian"
    },
    {
      "id": "pocket_tts_italian_bf16",
      "display_name": "PocketTTS Italian BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "PocketTTS-GGUF/italian",
      "files": [
        "PocketTTS-GGUF/italian/pocket-tts-italian-bf16.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/italian"
    },
    {
      "id": "pocket_tts_portuguese_q8_0",
      "display_name": "PocketTTS Portuguese Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "PocketTTS-GGUF/portuguese",
      "files": [
        "PocketTTS-GGUF/portuguese/pocket-tts-portuguese-q8_0.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/portuguese"
    },
    {
      "id": "pocket_tts_portuguese_bf16",
      "display_name": "PocketTTS Portuguese BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "PocketTTS-GGUF/portuguese",
      "files": [
        "PocketTTS-GGUF/portuguese/pocket-tts-portuguese-bf16.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/portuguese"
    },
    {
      "id": "pocket_tts_spanish_q8_0",
      "display_name": "PocketTTS Spanish Q8_0 GGUF",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "PocketTTS-GGUF/spanish",
      "files": [
        "PocketTTS-GGUF/spanish/pocket-tts-spanish-q8_0.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/spanish"
    },
    {
      "id": "pocket_tts_spanish_bf16",
      "display_name": "PocketTTS Spanish BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "PocketTTS-GGUF/spanish",
      "files": [
        "PocketTTS-GGUF/spanish/pocket-tts-spanish-bf16.gguf"
      ],
      "strip_prefix": "PocketTTS-GGUF/spanish"
    },
    {
      "id": "pocket_tts_english_safetensors",
      "display_name": "PocketTTS English Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "pocket-tts",
      "files": [
        "languages/english/embeddings/alba.safetensors",
        "languages/english/embeddings/anna.safetensors",
        "languages/english/embeddings/azelma.safetensors",
        "languages/english/embeddings/bill_boerst.safetensors",
        "languages/english/embeddings/caro_davy.safetensors",
        "languages/english/embeddings/charles.safetensors",
        "languages/english/embeddings/cosette.safetensors",
        "languages/english/embeddings/eponine.safetensors",
        "languages/english/embeddings/estelle.safetensors",
        "languages/english/embeddings/eve.safetensors",
        "languages/english/embeddings/fantine.safetensors",
        "languages/english/embeddings/george.safetensors",
        "languages/english/embeddings/giovanni.safetensors",
        "languages/english/embeddings/jane.safetensors",
        "languages/english/embeddings/javert.safetensors",
        "languages/english/embeddings/jean.safetensors",
        "languages/english/embeddings/juergen.safetensors",
        "languages/english/embeddings/lola.safetensors",
        "languages/english/embeddings/marius.safetensors",
        "languages/english/embeddings/mary.safetensors",
        "languages/english/embeddings/michael.safetensors",
        "languages/english/embeddings/paul.safetensors",
        "languages/english/embeddings/peter_yearsley.safetensors",
        "languages/english/embeddings/rafael.safetensors",
        "languages/english/embeddings/stuart_bell.safetensors",
        "languages/english/embeddings/vera.safetensors",
        "languages/english/model.safetensors",
        "languages/english/tokenizer.model"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "kyutai/pocket-tts"
      }
    }
  ]
}