Dynamic Accent Pro v1.1 (experimental prompt-controlled Nigerian TTS); Flash-DPO default, Pro via --instruct
582aab9 verified | { | |
| "default_version": "flash", | |
| "default_voice": "anchor_male", | |
| "versions": { | |
| "pro": { | |
| "engine": "cosyvoice3", | |
| "base": "FunAudioLLM/Fun-CosyVoice3-0.5B-2512", | |
| "lm": "checkpoints/pro/flash_instruct/best.pt", | |
| "diacritics": "checkpoints/flash/diacritics_byt5", | |
| "sampler": "silexcap", | |
| "one_pass": true, | |
| "postprocess": false, | |
| "rate": "normal", | |
| "frontend": "codeswitch", | |
| "translate": "experimental, off by default (--frontend translate --experimental_translate)", | |
| "desc": "Pro v1.1 (experimental; used automatically when --instruct is given): Flash-Instruct LM = Flash-DPO + instruction SFT (natural-language instruction in the CosyVoice3 system-prompt slot, audio-derived pace/pause labels, default prompt trained on native-like clips) + DPO polish (CER, pace/pause match to native, clipped-ending penalty); classifier-free guidance 2.0 on the instruction; few/many-pause instructions also edit the text ([pause] tags). Pace control works; pauses work through the text step; calm/lively and emotion are not supported. Licensed data only.", | |
| "instruct": true, | |
| "cfg": 2.0, | |
| "text_pauses": true, | |
| "tail_pad": 0 | |
| }, | |
| "flash": { | |
| "engine": "cosyvoice3", | |
| "base": "FunAudioLLM/Fun-CosyVoice3-0.5B-2512", | |
| "lm": "checkpoints/flash/flash_dpo/best.pt", | |
| "diacritics": "checkpoints/flash/diacritics_byt5", | |
| "sampler": "silexcap", | |
| "one_pass": true, | |
| "postprocess": false, | |
| "rate": "normal", | |
| "frontend": "codeswitch", | |
| "translate": "experimental, off by default (--frontend translate --experimental_translate)", | |
| "desc": "Flash (default without --instruct): Flash-DPO LM on Fun-CosyVoice3-0.5B-2512 (SFT on licensed data + DPO with a naturalness/CER/pace/pause reward), CV3 flow+vocoder zero-shot from the voice clip; text front-end restores Yoruba/Igbo diacritics; one pass per paragraph, model pauses, no post-processing" | |
| } | |
| }, | |
| "voices": [ | |
| "anchor_male", | |
| "waxal_ibo_5_female", | |
| "waxal_yor_5_female" | |
| ], | |
| "recommended": { | |
| "pro": [ | |
| "anchor_male", | |
| "waxal_ibo_5_female", | |
| "waxal_yor_5_female" | |
| ], | |
| "flash": [ | |
| "anchor_male", | |
| "waxal_ibo_5_female", | |
| "waxal_yor_5_female" | |
| ], | |
| "note": "Pro v1.1 is experimental. Pace instructions (slow/fast) work; pause instructions work through a text step; calm/lively has no measurable effect and is not advertised. --tail_pad 6 removes clipped endings (opt-in). Compare with --compare (flash vs pro) before use." | |
| }, | |
| "rates": [ | |
| "none", | |
| "very slow", | |
| "slow", | |
| "normal", | |
| "fast", | |
| "very fast" | |
| ] | |
| } |