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
351 lines
9.3 KiB
JSON
351 lines
9.3 KiB
JSON
{
|
|
"schema_version": 1,
|
|
"family": "ming3_tts",
|
|
"display_name": "Ming 3.0 TTS 0.5B",
|
|
"description": "Multilingual continuous-latent speech synthesis with a Qwen backbone, flow-matching acoustic head, CAMPPlus speaker conditioning and an audio VAE.",
|
|
"category": "tts",
|
|
"status": "supported",
|
|
"tasks": [
|
|
"tts",
|
|
"clone",
|
|
"design"
|
|
],
|
|
"modes": [
|
|
"offline",
|
|
"streaming"
|
|
],
|
|
"languages": [
|
|
"zh",
|
|
"en",
|
|
"ja",
|
|
"ko",
|
|
"de",
|
|
"fr",
|
|
"es"
|
|
],
|
|
"capabilities": {
|
|
"tts": [
|
|
"long_form",
|
|
"speaker_reference",
|
|
"style_control",
|
|
"emotion_control",
|
|
"built_in_voices"
|
|
],
|
|
"clone": [
|
|
"long_form",
|
|
"speaker_reference",
|
|
"style_control",
|
|
"emotion_control"
|
|
],
|
|
"design": [
|
|
"voice_design"
|
|
]
|
|
},
|
|
"runtime": {
|
|
"tags": [
|
|
"gguf"
|
|
]
|
|
},
|
|
"options": {
|
|
"request": [
|
|
{
|
|
"name": "template_name",
|
|
"type": "enum",
|
|
"required": false,
|
|
"description": "Conditioning template; defaults to voice_design on that route, otherwise tts.",
|
|
"values": [
|
|
"tts",
|
|
"multilingual",
|
|
"basic",
|
|
"emotion",
|
|
"dialect",
|
|
"ip",
|
|
"voice_design",
|
|
"freeform"
|
|
]
|
|
},
|
|
{
|
|
"name": "language",
|
|
"type": "string",
|
|
"required": false,
|
|
"description": "Target language name or code; defaults to the request's language (the server's top-level language field, the CLI's --language). Required by multilingual; ignored by dialect."
|
|
},
|
|
{
|
|
"name": "instruct",
|
|
"type": "string",
|
|
"required": false,
|
|
"description": "Voice description for voice_design; raw conditioning text for freeform or tts."
|
|
},
|
|
{
|
|
"name": "instruction_schema",
|
|
"type": "enum",
|
|
"required": false,
|
|
"description": "Structured instruction field language; defaults to en for dialect, otherwise zh.",
|
|
"values": [
|
|
"en",
|
|
"zh"
|
|
]
|
|
},
|
|
{
|
|
"name": "speed",
|
|
"type": "enum",
|
|
"required": false,
|
|
"description": "Categorical speaking speed for the basic template, not waveform resampling.",
|
|
"values": [
|
|
"慢速",
|
|
"中速",
|
|
"快速"
|
|
]
|
|
},
|
|
{
|
|
"name": "pitch",
|
|
"type": "enum",
|
|
"required": false,
|
|
"description": "Categorical pitch for the basic template.",
|
|
"values": [
|
|
"低",
|
|
"中",
|
|
"高"
|
|
]
|
|
},
|
|
{
|
|
"name": "volume",
|
|
"type": "enum",
|
|
"required": false,
|
|
"description": "Categorical volume for the basic template.",
|
|
"values": [
|
|
"低",
|
|
"中",
|
|
"高"
|
|
]
|
|
},
|
|
{
|
|
"name": "emotion",
|
|
"type": "string",
|
|
"required": false,
|
|
"description": "Emotion label for the emotion template."
|
|
},
|
|
{
|
|
"name": "dialect",
|
|
"type": "enum",
|
|
"required": false,
|
|
"description": "Dialect for the dialect template.",
|
|
"values": [
|
|
"chuanyu dialect",
|
|
"cantonese"
|
|
]
|
|
},
|
|
{
|
|
"name": "voice_id",
|
|
"type": "enum",
|
|
"required": false,
|
|
"description": "Built-in English voice for the ip template.",
|
|
"values": [
|
|
"US_Emma",
|
|
"US_James",
|
|
"UK_George",
|
|
"UK_Charlotte"
|
|
]
|
|
},
|
|
{
|
|
"name": "reference_text",
|
|
"type": "string",
|
|
"required": false,
|
|
"description": "Transcript accompanying voice_ref; enables the acoustic reference prefix. Without a transcript, only speaker conditioning is used."
|
|
},
|
|
{
|
|
"name": "speaker_embedding_file",
|
|
"type": "path_list",
|
|
"required": false,
|
|
"description": "CAMPPlus embedding file or list, with shape [192] or [1, 192]. Supersedes voice_ref audio; zero_speaker_embedding takes precedence."
|
|
},
|
|
{
|
|
"name": "export_speaker_embedding_file",
|
|
"type": "path",
|
|
"required": false,
|
|
"description": "Export the enabled voice_ref CAMPPlus speaker embedding as F32 Safetensors. Requires voice_ref, use_speaker_embedding=true, zero_speaker_embedding=false and no speaker_embedding_file. Does not export acoustic latents or the reference transcript."
|
|
},
|
|
{
|
|
"name": "use_speaker_embedding",
|
|
"type": "bool",
|
|
"required": false,
|
|
"default": true,
|
|
"description": "Condition on the reference speaker embedding."
|
|
},
|
|
{
|
|
"name": "zero_speaker_embedding",
|
|
"type": "bool",
|
|
"required": false,
|
|
"default": false,
|
|
"description": "Use a zero projected speaker row instead of reference audio conditioning."
|
|
},
|
|
{
|
|
"name": "max_tokens",
|
|
"type": "int",
|
|
"required": false,
|
|
"default": 1000,
|
|
"min": 1,
|
|
"description": "Maximum generated acoustic patches per text chunk, also limited by the 512-position generation context."
|
|
},
|
|
{
|
|
"name": "guidance_scale",
|
|
"type": "float",
|
|
"required": false,
|
|
"default": 2,
|
|
"description": "Flow guidance: conditioned prediction plus guidance_scale times its difference from the null prediction."
|
|
},
|
|
{
|
|
"name": "sigma",
|
|
"type": "float",
|
|
"required": false,
|
|
"default": 0.25,
|
|
"min": 0,
|
|
"description": "Flow SDE noise scale."
|
|
},
|
|
{
|
|
"name": "temperature",
|
|
"type": "float",
|
|
"required": false,
|
|
"default": 0,
|
|
"min": 0,
|
|
"description": "Flow sampling temperature; zero still uses seeded initial noise."
|
|
},
|
|
{
|
|
"name": "seed",
|
|
"type": "int",
|
|
"required": false,
|
|
"default": 1895,
|
|
"min": 0,
|
|
"description": "Reference-posterior and generation noise seed."
|
|
},
|
|
{
|
|
"name": "raw_text",
|
|
"type": "bool",
|
|
"required": false,
|
|
"default": false,
|
|
"description": "Send the text as one synthesis fragment without frontend chunking."
|
|
},
|
|
{
|
|
"name": "text_chunk_size",
|
|
"type": "int",
|
|
"required": false,
|
|
"min": 1,
|
|
"description": "Framework long-form chunk budget in Unicode codepoints; default 50 when hanzi, kana and Hangul outnumber other letters and digits, otherwise 300; multi-chunk generation without speaker conditioning is capped at 120, including explicit overrides, and anchors later chunks to the first generated voice. Smaller overrides, single-chunk inputs and explicitly conditioned voices retain their chunk size."
|
|
},
|
|
{
|
|
"name": "text_chunk_mode",
|
|
"type": "enum",
|
|
"required": false,
|
|
"default": "tag_aware",
|
|
"description": "Framework text chunking policy.",
|
|
"values": [
|
|
"default",
|
|
"tag_aware",
|
|
"japanese",
|
|
"endline"
|
|
]
|
|
},
|
|
{
|
|
"name": "trim_silence",
|
|
"type": "bool",
|
|
"required": false,
|
|
"default": true,
|
|
"description": "Trim trailing low-energy 100-ms frames per fragment, retaining a 300-ms tail."
|
|
},
|
|
{
|
|
"name": "stream_frames_per_event",
|
|
"type": "int",
|
|
"description": "Maximum latent patches per streaming audio decode.",
|
|
"required": false,
|
|
"default": 4,
|
|
"min": 1,
|
|
"max": 64
|
|
},
|
|
{
|
|
"name": "stream_chunk_policy",
|
|
"type": "enum",
|
|
"description": "Streaming audio chunk schedule.",
|
|
"required": false,
|
|
"default": "grow",
|
|
"values": [
|
|
"fixed",
|
|
"grow"
|
|
]
|
|
}
|
|
],
|
|
"session": [
|
|
{
|
|
"name": "reference_cache",
|
|
"type": "bool",
|
|
"required": false,
|
|
"default": true,
|
|
"description": "Reuse prepared conditioning for the most recently used reference waveform."
|
|
}
|
|
],
|
|
"load": []
|
|
},
|
|
"package_defaults": {
|
|
"download": {
|
|
"kind": "huggingface_snapshot",
|
|
"repo": "audio-cpp/Ming-3.0-TTS-0.5B-GGUF",
|
|
"revision": "main",
|
|
"gated": false
|
|
}
|
|
},
|
|
"packages": [
|
|
{
|
|
"id": "ming_3_0_tts_0_5b_bf16",
|
|
"display_name": "Ming 3.0 TTS 0.5B BF16",
|
|
"format": "gguf",
|
|
"precision": "bf16",
|
|
"default": true,
|
|
"target_directory": "Ming-3.0-TTS-0.5B-GGUF",
|
|
"files": [
|
|
"ming-3.0-tts-0.5b-bf16.gguf"
|
|
]
|
|
},
|
|
{
|
|
"id": "ming_3_0_tts_0_5b_q8_0",
|
|
"display_name": "Ming 3.0 TTS 0.5B Q8_0",
|
|
"format": "gguf",
|
|
"precision": "q8_0",
|
|
"default": false,
|
|
"target_directory": "Ming-3.0-TTS-0.5B-GGUF",
|
|
"files": [
|
|
"ming-3.0-tts-0.5b-q8_0.gguf"
|
|
]
|
|
}
|
|
],
|
|
"dependencies": [],
|
|
"ui": {
|
|
"recommended_package": "ming_3_0_tts_0_5b_bf16",
|
|
"tags": [
|
|
"TTS",
|
|
"GGUF"
|
|
],
|
|
"docs": [
|
|
"docs/models/ming3_tts.md"
|
|
]
|
|
},
|
|
"sources": [
|
|
{
|
|
"format": "gguf",
|
|
"roots": {
|
|
"model": ".",
|
|
"weights": "$gguf"
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"tokenizer_json": "model:tokenizer.json",
|
|
"tokenizer_config": "model:tokenizer_config.json"
|
|
},
|
|
"tensors": {
|
|
"weights": "weights:",
|
|
"speaker": {
|
|
"source": "weights:",
|
|
"prefix": "speaker"
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|