{ "host": "127.0.0.1", "port": 8088, "device": 0, "threads": 1, "_comment": "Whitelist of every model family currently enabled in registry.cpp. The WebUI loads ONE model at a time by (re)starting audiocpp_server with a single-model config; paths are relative to the bundle root (where cpu/ gpu/ models/ live). Entries whose directory is missing show as '未安装'; click 下载 in the UI to fetch the default GGUF package from model_specs/ in the background (runs `python webui/model_manager_webui.py install --models-root /models`). 'task' must be one of: tts, asr, vad, diar, sep, gen, clon, vc, s2s, align, vdes, spk, svc. UI tab mapping: tts/clon -> TTS 标签页, asr -> ASR, gen -> 音乐生成, vc/svc/s2s -> 声音转换, sep -> 音源分离, vad/diar/align -> 音频分析, vdes -> 声音设计. 'download_id' is the model_specs package id (omit for bundled assets like silero_vad). Optional per-entry keys: input_hint (overrides the family hint in webui.py), default_options, min_vram_gb. 'min_vram_gb' = ESTIMATED minimum CUDA VRAM (GB) to run typical requests at this entry's default precision/session_options — sources: docs/memory_saver.md official peak measurements (heartmula 25.6G@120s, stable-audio-medium 10.4G, chatterbox 13.4G, omnivoice 11.4G, qwen3-tts-1.7B 7.5G, stable-audio-small 3.7G), local RTX 4060 8G measurements (vibevoice 6.9G peak, ace-step w/ q8_0 preset), weight-size extrapolation for the rest. The UI warns when it exceeds detected local VRAM; exceeding means it may still run but will spill into shared memory and slow down badly.", "models": [ { "id": "omnivoice", "display_name": "OmniVoice (tts)", "family": "omnivoice", "path": "models/OmniVoice", "task": "tts", "mode": "offline", "download_id": "omnivoice", "min_vram_gb": 10 }, { "id": "pocket-tts", "display_name": "Pocket TTS (tts)", "family": "pocket_tts", "path": "models/pocket-tts", "task": "tts", "mode": "offline", "download_id": "pocket_tts", "min_vram_gb": 2 }, { "id": "qwen3-tts", "display_name": "Qwen3-TTS 0.6B (tts)", "family": "qwen3_tts", "path": "models/Qwen3-TTS-12Hz-0.6B-Base", "task": "tts", "mode": "offline", "download_id": "qwen3_tts_0_6b_base", "min_vram_gb": 5 }, { "id": "qwen3-tts-1.7b", "display_name": "Qwen3-TTS 1.7B Base (tts)", "family": "qwen3_tts", "path": "models/Qwen3-TTS-12Hz-1.7B-Base", "task": "tts", "mode": "offline", "download_id": "qwen3_tts_1_7b_base", "min_vram_gb": 8 }, { "id": "qwen3-tts-1.7b-custom", "display_name": "Qwen3-TTS 1.7B CustomVoice (tts)", "family": "qwen3_tts", "path": "models/Qwen3-TTS-12Hz-1.7B-CustomVoice", "task": "tts", "mode": "offline", "download_id": "qwen3_tts_1_7b_custom_voice", "min_vram_gb": 8 }, { "id": "miotts", "display_name": "MioTTS 1.7B (tts; needs MioCodec)", "family": "miotts", "path": "models/MioTTS-1.7B", "task": "tts", "mode": "offline", "download_id": "miotts_1_7b", "min_vram_gb": 8 }, { "id": "voxcpm2", "display_name": "VoxCPM2 (tts)", "family": "voxcpm2", "path": "models/VoxCPM2", "task": "tts", "mode": "offline", "download_id": "voxcpm2", "session_options": { "voxcpm2.weight_type": "q8_0" }, "min_vram_gb": 6 }, { "id": "vibevoice", "display_name": "VibeVoice 1.5B (tts, long-form/multi-speaker)", "family": "vibevoice", "path": "models/VibeVoice-1.5B", "task": "tts", "mode": "offline", "download_id": "vibevoice_1_5b", "min_vram_gb": 7 }, { "id": "index-tts2", "display_name": "IndexTTS2 (tts 中英克隆+情感)", "display_name_en": "IndexTTS2 (tts, zh/en clone + emotion)", "family": "index_tts2", "path": "models/IndexTTS-2", "task": "tts", "mode": "offline", "download_id": "index_tts2", "min_vram_gb": 8 }, { "id": "irodori-tts", "display_name": "Irodori-TTS 500M (tts 日语)", "display_name_en": "Irodori-TTS 500M (ja tts)", "family": "irodori_tts", "path": "models/Irodori-TTS-500M-v3", "task": "tts", "mode": "offline", "download_id": "irodori_tts_500m_v3", "min_vram_gb": 4 }, { "id": "moss-tts-local", "display_name": "MOSS-TTS-Local v1.5 (tts)", "family": "moss_tts_local", "path": "models/MOSS-TTS-Local-Transformer-v1.5", "task": "tts", "mode": "offline", "download_id": "moss_tts_local_v1_5", "min_vram_gb": 8 }, { "id": "moss-tts-nano", "display_name": "MOSS-TTS-Nano 100M (tts)", "family": "moss_tts_nano", "path": "models/MOSS-TTS-Nano-100M", "task": "tts", "mode": "offline", "download_id": "moss_tts_nano_100m", "min_vram_gb": 2 }, { "id": "supertonic", "display_name": "Supertonic 3 (tts 预置音色/多语种)", "display_name_en": "Supertonic 3 (tts, preset voices)", "family": "supertonic", "path": "models/supertonic-3", "task": "tts", "mode": "offline", "download_id": "supertonic_3", "min_vram_gb": 2 }, { "id": "higgs-audio-tts", "display_name": "Higgs Audio v3 TTS 4B (tts 克隆, GGUF Q8)", "display_name_en": "Higgs Audio v3 TTS 4B (tts + clone, GGUF Q8)", "family": "higgs_audio_tts", "path": "models/Higgs-Audio-v3-TTS-4B-GGUF", "task": "tts", "mode": "offline", "download_id": "higgs_audio_v3_tts_4b", "min_vram_gb": 6, "input_hint": "**Higgs Audio v3 TTS**：Q8_0 GGUF 包（权重已量化，不用再设 weight_type）；上传参考音色即声音克隆，留空用默认音色；长文本自动分段。", "input_hint_en": "**Higgs Audio v3 TTS**: Q8_0 GGUF package (already quantized — no weight_type needed). Upload a reference voice to clone, or leave it empty for the default voice; long text is chunked automatically." }, { "id": "fish-audio-s2-pro", "display_name": "Fish Audio S2 Pro (tts 克隆/控制标记, GGUF Q8)", "display_name_en": "Fish Audio S2 Pro (tts + clone/control tags, GGUF Q8)", "family": "fish_audio", "path": "models/Fish-Audio-S2-Pro-GGUF", "task": "tts", "mode": "offline", "download_id": "fish_audio_s2_pro", "min_vram_gb": 8, "input_hint": "**Fish Audio S2 Pro**：Q8_0 GGUF 包；中英+自动语种；上传参考音色即克隆；正文里可写行内控制标记（如 (laugh)）。", "input_hint_en": "**Fish Audio S2 Pro**: Q8_0 GGUF package; English/Chinese plus auto language. Upload a reference voice to clone; inline control tags such as (laugh) can be written in the text." }, { "id": "glm-tts", "display_name": "GLM-TTS (tts 克隆, 社区)", "display_name_en": "GLM-TTS (tts + clone, community)", "family": "glm_tts", "path": "models/GLM-TTS", "task": "tts", "mode": "offline", "download_id": "glm_tts", "min_vram_gb": 8, "input_hint": "**GLM-TTS**（社区模型）：中英 TTS / voice clone；上传参考音色即克隆。", "input_hint_en": "**GLM-TTS** (community): Chinese/English TTS and voice clone. Upload a reference voice to clone." }, { "id": "outetts", "display_name": "Llama-OuteTTS 1.0 1B (tts 克隆, 社区)", "display_name_en": "Llama-OuteTTS 1.0 1B (tts + clone, community)", "family": "outetts", "path": "models/Llama-OuteTTS-1.0-1B", "task": "tts", "mode": "offline", "download_id": "outetts_1_0_1b", "min_vram_gb": 4, "input_hint": "**OuteTTS 1.0 1B**（社区模型）：23 种语言，DAC 编解码；上传参考音色即克隆。", "input_hint_en": "**OuteTTS 1.0 1B** (community): 23 languages, IBM DAC codec. Upload a reference voice to clone." }, { "id": "vietneu-tts", "display_name": "VieNeu-TTS v3 Turbo (tts 越南语, 社区)", "display_name_en": "VieNeu-TTS v3 Turbo (vi tts, community)", "family": "vietneu_tts", "path": "models/VieNeu-TTS-v3-Turbo", "task": "tts", "mode": "offline", "download_id": "vietneu_tts_v3_turbo", "min_vram_gb": 4, "input_hint": "**VieNeu-TTS v3 Turbo**（社区模型）：越南语 / 英语；上传参考音色即克隆。", "input_hint_en": "**VieNeu-TTS v3 Turbo** (community): Vietnamese and English. Upload a reference voice to clone." }, { "id": "inflect-v2", "display_name": "Inflect Micro v2 (tts 英语, 社区)", "display_name_en": "Inflect Micro v2 (en tts, community)", "family": "inflect_v2", "path": "models/Inflect-Micro-v2", "task": "tts", "mode": "offline", "download_id": "inflect_micro_v2", "min_vram_gb": 2, "input_hint": "**Inflect Micro v2**（社区模型）：英语离线 TTS；Micro 是默认包，Nano 可通过模型管理器另装后手动选择路径。", "input_hint_en": "**Inflect Micro v2** (community): English offline TTS. Micro is the default package; Nano can be installed separately and selected manually." }, { "id": "dramabox", "display_name": "DramaBox (tts 克隆, GGUF Q8)", "display_name_en": "DramaBox (tts + clone, GGUF Q8)", "family": "dramabox", "path": "models/DramaBox-GGUF", "task": "tts", "mode": "offline", "download_id": "dramabox_q8_0", "min_vram_gb": 16, "input_hint": "**DramaBox**：英语 TTS / voice clone；上传参考音色可克隆，长文本建议写清楚说话人描述。", "input_hint_en": "**DramaBox**: English TTS and voice clone. Upload a reference voice to clone; for long text, keep speaker wording explicit." }, { "id": "confucius4-tts", "display_name": "Confucius4-TTS (voice clone, GGUF)", "display_name_en": "Confucius4-TTS (voice clone, GGUF)", "family": "confucius4_tts", "path": "models/Confucius4-TTS-GGUF", "task": "clon", "mode": "offline", "download_id": "confucius4_tts_orig", "min_vram_gb": 8, "input_hint": "**Confucius4-TTS**：需要参考音色；当前中文/英语路径更可靠，非中英语种仍在验证中。", "input_hint_en": "**Confucius4-TTS**: requires a reference voice. Chinese/English are the most reliable paths; other languages are still being validated." }, { "id": "chatterbox", "display_name": "Chatterbox (voice clone)", "family": "chatterbox", "path": "models/chatterbox", "task": "clon", "mode": "offline", "download_id": "chatterbox", "min_vram_gb": 12 }, { "id": "ace-step", "display_name": "ACE-Step 1.5 (music gen)", "family": "ace_step", "path": "models/Ace-Step1.5", "task": "gen", "mode": "offline", "download_id": "ace_step", "session_options": { "ace_step.mem_saver": "true", "ace_step.dit_weight_type": "q8_0", "ace_step.text_encoder_weight_type": "q8_0", "ace_step.planner_weight_type": "q8_0" }, "min_vram_gb": 8 }, { "id": "stable-audio-small-music","display_name": "Stable Audio 3 Small Music (gen)", "family": "stable_audio", "path": "models/stable-audio-3-small-music", "task": "gen", "mode": "offline", "download_id": "stable_audio_3_small_music", "min_vram_gb": 4 }, { "id": "stable-audio-small-sfx", "display_name": "Stable Audio 3 Small SFX (gen)", "family": "stable_audio", "path": "models/stable-audio-3-small-sfx", "task": "gen", "mode": "offline", "download_id": "stable_audio_3_small_sfx", "min_vram_gb": 4 }, { "id": "stable-audio-medium", "display_name": "Stable Audio 3 Medium (gen)", "family": "stable_audio", "path": "models/stable-audio-3-medium", "task": "gen", "mode": "offline", "download_id": "stable_audio_3_medium", "session_options": { "stable_audio.mem_saver": "true" }, "min_vram_gb": 10 }, { "id": "heartmula", "display_name": "HeartMuLa 3B (music gen)", "family": "heartmula", "path": "models/HeartMuLa", "task": "gen", "mode": "offline", "download_id": "heartmula", "session_options": { "heartmula.mem_saver": "true" }, "min_vram_gb": 24 }, { "id": "qwen3-asr", "display_name": "Qwen3-ASR 0.6B (asr)", "family": "qwen3_asr", "path": "models/Qwen3-ASR-0.6B", "task": "asr", "mode": "offline", "download_id": "qwen3_asr_0_6b", "min_vram_gb": 3 }, { "id": "qwen3-asr-1.7b", "display_name": "Qwen3-ASR 1.7B HF (asr)", "family": "qwen3_asr", "path": "models/Qwen3-ASR-1.7B-hf", "task": "asr", "mode": "offline", "download_id": "qwen3_asr_1_7b_hf", "min_vram_gb": 6, "input_hint": "**Qwen3-ASR 1.7B**（HF 原生权重，免转换）：精度高于 0.6B；长音频自动分段转写；8G 卡显存偏紧，长音频建议先短段试跑。" }, { "id": "citrinet-asr", "display_name": "Citrinet ASR (asr)", "family": "citrinet_asr", "path": "models/citrinet", "task": "asr", "mode": "offline", "download_id": "citrinet_asr", "min_vram_gb": 2 }, { "id": "nemotron-asr", "display_name": "Nemotron 3.5 ASR 0.6B (asr, 100+语种)", "family": "nemotron_asr", "path": "models/nemotron-3.5-asr-streaming-0.6b", "task": "asr", "mode": "offline", "download_id": "nemotron_asr", "min_vram_gb": 4, "input_hint": "**Nemotron ASR**：100+ 语种，语种码为 BCP-47（如 en-US / zh-CN），留空=auto；模型自带长音频处理。" }, { "id": "higgs-audio-stt", "display_name": "Higgs Audio v3 STT (asr, 英语)", "family": "higgs_audio_stt", "path": "models/higgs-audio-v3-stt", "task": "asr", "mode": "offline", "download_id": "higgs_audio_stt", "min_vram_gb": 8, "input_hint": "**Higgs Audio STT**：英语转写；可在文本框填指令（默认相当于 Transcribe the speech.）；离线模式自动切分长音频。" }, { "id": "hviske-asr", "display_name": "Hviske v5.3 (asr, 丹麦语)", "family": "hviske_asr", "path": "models/hviske-v5.3", "task": "asr", "mode": "offline", "download_id": "hviske_asr", "min_vram_gb": 6, "input_hint": "**Hviske ASR**：丹麦语专用；模型侧自动分段。" }, { "id": "vibevoice-asr", "display_name": "VibeVoice ASR (asr, 多语种+说话人分段)", "family": "vibevoice_asr", "path": "models/VibeVoice-ASR", "task": "asr", "mode": "offline", "download_id": "vibevoice_asr", "min_vram_gb": 20, "input_hint": "**VibeVoice ASR**：自动语种，可输出分段/说话人轮次；文本框可填上下文提示（如 The recording is a meeting conversation.）。权重 17.3G，8G 卡跑不动。" }, { "id": "voxtral-realtime", "display_name": "Voxtral Mini 4B Realtime (asr, 自动语种+流式)", "display_name_en": "Voxtral Mini 4B Realtime (asr, auto + streaming)", "family": "voxtral_realtime", "path": "models/Voxtral-Mini-4B-Realtime-2602-GGUF", "task": "asr", "mode": "offline", "download_id": "voxtral_realtime", "min_vram_gb": 8 }, { "id": "fun-asr-nano", "display_name": "Fun-ASR-Nano 2512 (asr, GGUF Q8)", "display_name_en": "Fun-ASR-Nano 2512 (asr, GGUF Q8)", "family": "fun_asr_nano", "path": "models/Fun-ASR-Nano-2512-GGUF", "task": "asr", "mode": "offline", "download_id": "fun_asr_nano_2512_q8_0", "min_vram_gb": 4, "input_hint": "**Fun-ASR-Nano**：轻量离线 ASR；支持 auto/中文/英语/日语。", "input_hint_en": "**Fun-ASR-Nano**: lightweight offline ASR; supports auto, Chinese, English and Japanese." }, { "id": "parakeet-tdt", "display_name": "Parakeet-TDT 0.6B v3 (asr, 流式)", "display_name_en": "Parakeet-TDT 0.6B v3 (asr + streaming)", "family": "parakeet_tdt", "path": "models/parakeet-tdt-0.6b-v3", "task": "asr", "mode": "offline", "download_id": "parakeet_tdt", "min_vram_gb": 4, "input_hint": "**Parakeet-TDT**：离线/长音频/流式 ASR；支持多种欧洲语言，留空=自动。", "input_hint_en": "**Parakeet-TDT**: offline, long-form and streaming ASR for many European languages; leave language empty for auto." }, { "id": "kroko-asr", "display_name": "Kroko Community ASR (asr, GGUF Q8)", "display_name_en": "Kroko Community ASR (asr, GGUF Q8)", "family": "kroko_asr", "path": "models/Kroko-ASR-GGUF", "task": "asr", "mode": "offline", "download_id": "kroko_asr_community_q8_0", "min_vram_gb": 4, "input_hint": "**Kroko Community ASR**：GGUF Q8 包；离线转写，支持时间戳。", "input_hint_en": "**Kroko Community ASR**: GGUF Q8 package for offline transcription with timestamps." }, { "id": "chatterbox-vc", "display_name": "Chatterbox (vc 声音转换)", "display_name_en": "Chatterbox (voice conversion)", "family": "chatterbox", "path": "models/chatterbox", "task": "vc", "mode": "offline", "download_id": "chatterbox", "min_vram_gb": 12, "input_hint": "**Chatterbox VC**：上传源语音和目标音色参考；模型保留源语音内容，将说话人音色转换为目标音色，输出 24kHz 单声道。", "input_hint_en": "**Chatterbox VC**: upload source speech and a target-voice reference. It preserves the source content and converts the speaker identity; output is 24 kHz mono." }, { "id": "vevo2", "display_name": "Vevo2 (vc 语音转换, GGUF Q8)", "display_name_en": "Vevo2 (voice conversion, GGUF Q8)", "family": "vevo2", "path": "models/Vevo2-GGUF", "task": "vc", "mode": "offline", "download_id": "vevo2_gguf", "min_vram_gb": 6 }, { "id": "vevo2-svc", "display_name": "Vevo2 (svc 歌声转换, GGUF Q8)", "display_name_en": "Vevo2 (singing voice conversion, GGUF Q8)", "family": "vevo2", "path": "models/Vevo2-GGUF", "task": "svc", "mode": "offline", "download_id": "vevo2_gguf", "min_vram_gb": 6, "input_hint": "**Vevo2 歌声转换 (svc)**：上传源歌声 + 目标歌手参考音色，默认 route=style_preserved_svc。style_converted_svc / singing_style_conversion 等风格转换 route 需在『其它参数(JSON)』里补 `style_ref`（服务器本地 wav 路径）/ `style_ref_text` / `target_text`。" }, { "id": "vevo2-s2s", "display_name": "Vevo2 (s2s 语音编辑, GGUF Q8)", "display_name_en": "Vevo2 (speech editing, GGUF Q8)", "family": "vevo2", "path": "models/Vevo2-GGUF", "task": "s2s", "mode": "offline", "download_id": "vevo2_gguf", "min_vram_gb": 6, "input_hint": "**Vevo2 语音编辑 (s2s)**：上传要编辑的源语音，并在『其它参数(JSON)』里填 `{\"target_text\": \"替换后的完整句子\"}`（编辑保持原说话人音色，可不上传目标音色）。" }, { "id": "seed-vc", "display_name": "Seed-VC (vc 语音转换)", "family": "seed_vc", "path": "models/SeedVC-MLX", "task": "vc", "mode": "offline", "download_id": "seed_vc", "min_vram_gb": 4 }, { "id": "seed-vc-svc", "display_name": "Seed-VC (svc 歌声转换)", "family": "seed_vc", "path": "models/SeedVC-MLX", "task": "svc", "mode": "offline", "download_id": "seed_vc", "min_vram_gb": 4, "input_hint": "**Seed-VC 歌声转换 (svc)**：上传源歌声 + 目标歌手参考音色，默认 route=v1_svc（带 F0 条件）。可在『其它参数(JSON)』里调 `auto_f0_adjust` / `semi_tone_shift` / `f0_condition`。" }, { "id": "rvc", "display_name": "RVC (vc, GGUF F16)", "display_name_en": "RVC (voice conversion, GGUF F16)", "family": "rvc", "path": "models/RVC-GGUF", "task": "vc", "mode": "offline", "download_id": "rvc_f16", "min_vram_gb": 4, "input_hint": "**RVC**：所选 GGUF 即目标音色；上传源语音即可转换；索引/音高等选项可用 JSON 传。", "input_hint_en": "**RVC**: the selected GGUF is the target voice; upload source speech to convert. Index and pitch options can be passed through the JSON box." }, { "id": "miocodec", "display_name": "MioCodec (vc; codec dependency)", "family": "miocodec", "path": "models/MioCodec-25Hz-44.1kHz-v2", "task": "vc", "mode": "offline", "download_id": "miocodec_25hz_44k_v2", "min_vram_gb": 3 }, { "id": "htdemucs", "display_name": "HTDemucs (sep 音源分离)", "family": "htdemucs", "path": "models/htdemucs", "task": "sep", "mode": "offline", "download_id": "htdemucs", "min_vram_gb": 3 }, { "id": "bs-roformer", "display_name": "BS-RoFormer (sep 人声分离)", "family": "bs_roformer", "path": "models/BS-RoFormer-ep368_Q8/BS-RoFormer-ep368_Q8.gguf", "task": "sep", "mode": "offline", "min_vram_gb": 3 }, { "id": "mel-band-roformer", "display_name": "Mel-Band RoFormer (sep 人声分离)", "family": "mel_band_roformer", "path": "models/mel-roformer-mlx", "task": "sep", "mode": "offline", "download_id": "mel_band_roformer", "min_vram_gb": 3 }, { "id": "silero-vad", "display_name": "Silero VAD (vad, bundled)", "family": "silero_vad", "path": "assets/framework/models/silero_vad", "task": "vad", "mode": "offline", "min_vram_gb": 1 }, { "id": "marblenet-vad", "display_name": "MarbleNet VAD (vad, bundled)", "family": "marblenet_vad", "path": "assets/framework/models/marblenet_vad", "task": "vad", "mode": "offline", "min_vram_gb": 1 }, { "id": "sortformer-diar", "display_name": "Sortformer Diarization 4spk (diar)", "family": "sortformer_diar", "path": "models/diar_sortformer_4spk-v1", "task": "diar", "mode": "offline", "download_id": "sortformer_diar_4spk_v1", "min_vram_gb": 2 }, { "id": "qwen3-forced-aligner", "display_name": "Qwen3 Forced Aligner (align)", "family": "qwen3_forced_aligner", "path": "models/Qwen3-ForcedAligner-0.6B", "task": "align", "mode": "offline", "download_id": "qwen3_forced_aligner_0_6b", "min_vram_gb": 3 }, { "id": "qwen3-tts-1.7b-vdesign", "display_name": "Qwen3-TTS 1.7B VoiceDesign (vdes)", "family": "qwen3_tts", "path": "models/Qwen3-TTS-12Hz-1.7B-VoiceDesign", "task": "vdes", "mode": "offline", "download_id": "qwen3_tts_1_7b_voice_design", "min_vram_gb": 8, "input_hint": "**Qwen3-TTS VoiceDesign**：在『音色描述』里用文字描述想要的声音（如“低沉磁性的中年男声，语速偏慢”），配上要念的文本即可，无需参考音频。" }, { "id": "irodori-tts-vdesign", "display_name": "Irodori-TTS 600M VoiceDesign (vdes 日语)", "display_name_en": "Irodori-TTS 600M VoiceDesign (ja vdes)", "family": "irodori_tts", "path": "models/Irodori-TTS-600M-v3-VoiceDesign", "task": "vdes", "mode": "offline", "download_id": "irodori_tts_600m_v3_voice_design", "min_vram_gb": 4, "input_hint": "**Irodori-TTS VoiceDesign**（日语）：『音色描述』用日语 caption 描述音色（如「落ち着いた大人の男性。深く響く声。」），文本填要念的日语内容，无需参考音频。" } ] }