Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
424 lines
12 KiB
JSON
424 lines
12 KiB
JSON
{
|
|
"schema_version": 1,
|
|
"family": "dots_tts",
|
|
"display_name": "DotTTS",
|
|
"description": "DotTTS is a multilingual zero-shot TTS, voice-cloning, and speech-editing model family packaged for audio.cpp with SOAR and MeanFlow generation, prompt-audio plus prompt-text prefill, speaker-reference conditioning, language tags, multiple synthesis templates, long-form text chunking, and true streaming audio output.",
|
|
"category": "tts",
|
|
"status": "supported",
|
|
"tasks": [
|
|
"tts",
|
|
"clone"
|
|
],
|
|
"modes": [
|
|
"offline",
|
|
"streaming"
|
|
],
|
|
"languages": [
|
|
"multilingual"
|
|
],
|
|
"runtime": {
|
|
"tags": [
|
|
"gguf",
|
|
"stream"
|
|
]
|
|
},
|
|
"capabilities": {
|
|
"tts": [
|
|
"speaker_reference",
|
|
"style_control",
|
|
"long_form"
|
|
],
|
|
"clone": [
|
|
"speaker_reference",
|
|
"style_control",
|
|
"long_form"
|
|
]
|
|
},
|
|
"options": {
|
|
"request": [
|
|
{
|
|
"name": "reference_text",
|
|
"type": "string",
|
|
"description": "Transcript for the prompt audio. When set with reference audio, DotTTS uses prompt prefill for higher quality voice cloning.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "reference_duration_sec",
|
|
"type": "float",
|
|
"description": "Optional prompt-audio trim duration before reference conditioning and prompt prefill.",
|
|
"required": false,
|
|
"min": 0.0
|
|
},
|
|
{
|
|
"name": "template_name",
|
|
"type": "enum",
|
|
"description": "DotTTS synthesis template. tts is normal text-to-speech; instruction_tts adds instruction conditioning; text_to_audio uses a sound-description template; tts_interleave uses the streaming interleave schedule; edit uses the DotTTS Edit source-audio editing schedule and is offline-only.",
|
|
"values": [
|
|
"tts",
|
|
"instruction_tts",
|
|
"text_to_audio",
|
|
"tts_interleave",
|
|
"edit"
|
|
],
|
|
"required": false,
|
|
"default": "tts"
|
|
},
|
|
{
|
|
"name": "source_audio",
|
|
"type": "audio_path",
|
|
"description": "Source WAV path for DotTTS Edit. Required when template_name=edit unless audio_input is supplied by the caller.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "instruct",
|
|
"type": "string",
|
|
"description": "Structured DotTTS Edit instruction. If omitted with template_name=edit, the request text is used as the instruction.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "source_text",
|
|
"type": "string",
|
|
"description": "Optional source transcript override for DotTTS Edit. If omitted, it is rendered from instruction tags.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "target_text",
|
|
"type": "string",
|
|
"description": "Edit target transcript override. When omitted, explicit instruction plus text input uses that text; otherwise the target is rendered from instruction tags.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "use_xvector",
|
|
"type": "enum",
|
|
"description": "DotTTS Edit speaker conditioning mode. auto follows the upstream tag-based behavior; on/off force or disable source-audio speaker conditioning.",
|
|
"values": [
|
|
"auto",
|
|
"on",
|
|
"off"
|
|
],
|
|
"required": false,
|
|
"default": "auto"
|
|
},
|
|
{
|
|
"name": "language",
|
|
"type": "string",
|
|
"description": "Optional target language tag. none disables language tagging; otherwise pass a language code such as en or zh.",
|
|
"required": false,
|
|
"default": "none"
|
|
},
|
|
{
|
|
"name": "num_inference_steps",
|
|
"type": "int",
|
|
"description": "SOAR flow-matching inference step count; default 10.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 10
|
|
},
|
|
{
|
|
"name": "guidance_scale",
|
|
"type": "float",
|
|
"description": "Classifier-free guidance scale used by SOAR flow matching. Default 1.2.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 1.2
|
|
},
|
|
{
|
|
"name": "speaker_scale",
|
|
"type": "float",
|
|
"description": "Multiplier applied to the prompt speaker embedding; default 1.5.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 1.5
|
|
},
|
|
{
|
|
"name": "sampler_mode",
|
|
"type": "enum",
|
|
"description": "SOAR ODE solver method for flow matching. Fixed-step euler is the default and parity baseline.",
|
|
"values": [
|
|
"euler",
|
|
"midpoint",
|
|
"rk4"
|
|
],
|
|
"required": false,
|
|
"default": "euler"
|
|
},
|
|
{
|
|
"name": "max_tokens",
|
|
"type": "int",
|
|
"description": "Maximum generated audio patch count per generated segment; default 500. In Edit mode source patches consume this budget; it must exceed the source-patch count.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 500
|
|
},
|
|
{
|
|
"name": "text_chunk_size",
|
|
"type": "int",
|
|
"description": "Maximum text codepoints per generated segment for long-form requests; default 320.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 320
|
|
},
|
|
{
|
|
"name": "text_chunk_mode",
|
|
"type": "enum",
|
|
"description": "Framework text chunking mode; default tag_aware preserves leading style/control tags across long-form chunks.",
|
|
"values": [
|
|
"default",
|
|
"tag_aware",
|
|
"japanese",
|
|
"endline"
|
|
],
|
|
"required": false,
|
|
"default": "tag_aware"
|
|
},
|
|
{
|
|
"name": "vocoder_merge_steps",
|
|
"type": "int",
|
|
"description": "Streaming vocoder latent patch merge size after the initial unmerged patches; default 4, matching the Python optimized streaming runtime default.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 4
|
|
},
|
|
{
|
|
"name": "seed",
|
|
"type": "int",
|
|
"description": "Seed for prompt latent sampling and flow noise initialization; default 42.",
|
|
"required": false,
|
|
"min": 0,
|
|
"default": 42
|
|
}
|
|
],
|
|
"session": [
|
|
{
|
|
"name": "weight_type",
|
|
"type": "enum",
|
|
"description": "Shared matmul weight storage type for DotTTS components; default native. Component-specific weight options override this value.",
|
|
"preset": "weight_type_full",
|
|
"required": false,
|
|
"default": "native"
|
|
},
|
|
{
|
|
"name": "speaker_encoder_weight_type",
|
|
"type": "enum",
|
|
"description": "Speaker encoder matmul weight storage type; defaults to weight_type.",
|
|
"preset": "weight_type_full",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "codec_weight_type",
|
|
"type": "enum",
|
|
"description": "AudioVAE vocoder matmul and recurrent weight storage type; defaults to weight_type.",
|
|
"preset": "weight_type_full",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "patch_encoder_weight_type",
|
|
"type": "enum",
|
|
"description": "Patch encoder transformer matmul weight storage type; defaults to weight_type.",
|
|
"preset": "weight_type_full",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "llm_weight_type",
|
|
"type": "enum",
|
|
"description": "Autoregressive language model matmul weight storage type; defaults to weight_type.",
|
|
"preset": "weight_type_full",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "flow_weight_type",
|
|
"type": "enum",
|
|
"description": "SOAR or MeanFlow DiT flow-matching matmul weight storage type; defaults to weight_type.",
|
|
"preset": "weight_type_full",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "codec_conv_weight_type",
|
|
"type": "enum",
|
|
"description": "Convolution weight storage type; default native.",
|
|
"preset": "weight_type_conv",
|
|
"required": false,
|
|
"default": "native"
|
|
},
|
|
{
|
|
"name": "reference_cache_slots",
|
|
"type": "int",
|
|
"description": "Prepared reference-audio cache slots; set 0 to disable reuse.",
|
|
"required": false,
|
|
"min": 0,
|
|
"default": 4
|
|
},
|
|
{
|
|
"name": "mem_saver",
|
|
"type": "bool",
|
|
"description": "Load DotTTS components by request phase and release their weights/graphs after last use while keeping reference cache slots alive.",
|
|
"required": false,
|
|
"default": false
|
|
}
|
|
],
|
|
"load": []
|
|
},
|
|
"package_defaults": {
|
|
"download": {
|
|
"kind": "huggingface_snapshot",
|
|
"repo": "audio-cpp/audio.cpp-gguf",
|
|
"revision": "main",
|
|
"gated": false
|
|
}
|
|
},
|
|
"packages": [
|
|
{
|
|
"id": "dots_tts_soar_q8_0",
|
|
"display_name": "DotTTS SOAR Q8_0 GGUF",
|
|
"default": true,
|
|
"format": "gguf",
|
|
"precision": "q8_0",
|
|
"target_directory": "DotTTS-SOAR-GGUF",
|
|
"files": [
|
|
"DotTTS-SOAR-GGUF/dots-tts-soar-q8_0.gguf"
|
|
],
|
|
"strip_prefix": "DotTTS-SOAR-GGUF"
|
|
},
|
|
{
|
|
"id": "dots_tts_soar_bf16",
|
|
"display_name": "DotTTS SOAR BF16 GGUF",
|
|
"default": false,
|
|
"format": "gguf",
|
|
"precision": "bf16",
|
|
"target_directory": "DotTTS-SOAR-GGUF",
|
|
"files": [
|
|
"DotTTS-SOAR-GGUF/dots-tts-soar-bf16.gguf"
|
|
],
|
|
"strip_prefix": "DotTTS-SOAR-GGUF"
|
|
},
|
|
{
|
|
"id": "dots_tts_mf_q8_0",
|
|
"display_name": "DotTTS MeanFlow Q8_0 GGUF",
|
|
"default": false,
|
|
"format": "gguf",
|
|
"precision": "q8_0",
|
|
"target_directory": "DotTTS-MF-GGUF",
|
|
"files": [
|
|
"DotTTS-MF-GGUF/dots-tts-mf-q8_0.gguf"
|
|
],
|
|
"strip_prefix": "DotTTS-MF-GGUF"
|
|
},
|
|
{
|
|
"id": "dots_tts_mf_bf16",
|
|
"display_name": "DotTTS MeanFlow BF16 GGUF",
|
|
"default": false,
|
|
"format": "gguf",
|
|
"precision": "bf16",
|
|
"target_directory": "DotTTS-MF-GGUF",
|
|
"files": [
|
|
"DotTTS-MF-GGUF/dots-tts-mf-bf16.gguf"
|
|
],
|
|
"strip_prefix": "DotTTS-MF-GGUF"
|
|
},
|
|
{
|
|
"id": "dots_tts_edit_q8_0",
|
|
"display_name": "DotTTS Edit Q8_0 GGUF",
|
|
"default": false,
|
|
"format": "gguf",
|
|
"precision": "q8_0",
|
|
"target_directory": "DotTTS-Edit-GGUF",
|
|
"files": [
|
|
"DotTTS-Edit-GGUF/dots-tts-edit-q8_0.gguf"
|
|
],
|
|
"strip_prefix": "DotTTS-Edit-GGUF"
|
|
},
|
|
{
|
|
"id": "dots_tts_edit_bf16",
|
|
"display_name": "DotTTS Edit BF16 GGUF",
|
|
"default": false,
|
|
"format": "gguf",
|
|
"precision": "bf16",
|
|
"target_directory": "DotTTS-Edit-GGUF",
|
|
"files": [
|
|
"DotTTS-Edit-GGUF/dots-tts-edit-bf16.gguf"
|
|
],
|
|
"strip_prefix": "DotTTS-Edit-GGUF"
|
|
}
|
|
],
|
|
"dependencies": [],
|
|
"ui": {
|
|
"recommended_package": "dots_tts_soar_q8_0",
|
|
"tags": [
|
|
"TTS",
|
|
"Clone",
|
|
"Stream",
|
|
"GGUF"
|
|
],
|
|
"docs": [
|
|
"docs/tts.md",
|
|
"docs/gguf.md"
|
|
]
|
|
},
|
|
"sources": [
|
|
{
|
|
"format": "gguf",
|
|
"roots": {
|
|
"model": ".",
|
|
"weights": "$gguf"
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"llm_config": "model:llm_config.json",
|
|
"tokenizer_json": "model:tokenizer.json",
|
|
"tokenizer_config": "model:tokenizer_config.json",
|
|
"vocab": "model:vocab.json",
|
|
"merges": "model:merges.txt",
|
|
"added_tokens": "model:added_tokens.json",
|
|
"special_tokens_map": "model:special_tokens_map.json"
|
|
},
|
|
"tensors": {
|
|
"core_weights": {
|
|
"source": "weights:",
|
|
"prefix": "core"
|
|
},
|
|
"vocoder_weights": {
|
|
"source": "weights:",
|
|
"prefix": "vocoder"
|
|
},
|
|
"speaker_encoder_weights": {
|
|
"source": "weights:",
|
|
"prefix": "speaker_encoder"
|
|
},
|
|
"latent_stats": {
|
|
"source": "weights:",
|
|
"prefix": "latent_stats"
|
|
}
|
|
}
|
|
},
|
|
{
|
|
"format": "safetensors",
|
|
"roots": {
|
|
"model": "."
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"llm_config": "model:llm_config.json",
|
|
"tokenizer_json": "model:tokenizer.json",
|
|
"tokenizer_config": "model:tokenizer_config.json",
|
|
"vocab": "model:vocab.json",
|
|
"merges": "model:merges.txt",
|
|
"added_tokens": "model:added_tokens.json",
|
|
"special_tokens_map": "model:special_tokens_map.json"
|
|
},
|
|
"tensors": {
|
|
"core_weights": {
|
|
"source": "model:model.safetensors"
|
|
},
|
|
"vocoder_weights": {
|
|
"source": "model:vocoder.safetensors"
|
|
},
|
|
"speaker_encoder_weights": {
|
|
"source": "model:speaker_encoder.safetensors"
|
|
},
|
|
"latent_stats": {
|
|
"source": "model:latent_stats.safetensors"
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|