Commit Graph
279 Commits
Author SHA1 Message Date
Hawk Lee 51a781ba2f feat: convert guidance inputs to static text labels using custom JS extension 2025-12-31 10:28:36 +08:00
Hawk Lee 52afa11e25 fix: implement robust model switching logic and fix instruction hallucination 2025-12-31 10:15:49 +08:00
Hawk Lee b5851715d2 style: add explanatory text labels above TTS input fields for better UI/UX 2025-12-31 10:04:04 +08:00
Hawk Lee 8c674f8c43 fix: prioritize RL-Instruct model and fix instruction hallucination issues 2025-12-31 01:12:55 +08:00
Hawk Lee 3ae14ba127 chore: suppress noisy PyTorch and ONNX warnings in terminal output 2025-12-31 01:01:22 +08:00
Hawk Lee 30af75d8bc feat: enable Pure Instruct Mode for CosyVoice 3.0 via internal neutral seed 2025-12-31 00:57:10 +08:00
Hawk Lee 12e589394e fix: resolve KeyError 'pure_1' in CosyVoice 3.0 by adding speaker validation and better defaults 2025-12-31 00:34:08 +08:00
Hawk Lee b217dcee89 feat: integrate CosyVoice 3.0 with full capabilities (Instruct, Zero-Shot, Hybrid, SFT, Cross-Lingual) and update documentation 2025-12-31 00:23:40 +08:00
Hawk Lee 6ca15fc123 Revert "Standard TTS: Add 'voice' selector and reference audio file support"
This reverts commit 2523dd6ad8.
2025-12-30 23:42:08 +08:00
Hawk Lee 2523dd6ad8 Standard TTS: Add 'voice' selector and reference audio file support 2025-12-30 23:40:28 +08:00
Hawk Lee 73bad5abe7 Fix: Remove resilient debug print 'Full Injection' 2025-12-30 23:30:54 +08:00
Hawk Lee 6648f3c655 Optimize 0.5B integration: Cleanup debug prints, update docs, mark Preset Maker as experimental 2025-12-30 23:25:39 +08:00
Hawk Lee 6614ac07cf Refactor: Separate VibeVoice Realtime TTS and Preset Maker into modular files and enable direct connection 2025-12-30 16:32:30 +08:00
Hawk Lee 8cf32887d1 feat: integrate VibeVoice 0.5B realtime support and fix JS deprecation warnings 2025-12-29 20:43:02 +08:00
Hawk Lee d4f560cdbb Add support for VibeVoice-Realtime-0.5B (Multi-language & Low latency) 2025-12-29 19:29:21 +08:00
Hawk Lee bcedf6b700 Introduce 'auto' sampling mode to fix 1.5B quality while keeping 7B expressiveness 2025-12-29 19:25:35 +08:00
Hawk Lee 81f0112094 Smooth progress bar with text-length estimation and final 7B quality fixes 2025-12-29 19:21:34 +08:00
Hawk Lee 2ac1d61790 Implement model-specific generation config loading (1.5B vs 7B presets) 2025-12-29 19:15:20 +08:00
Hawk Lee ce5182499b Sync defaults with official Microsoft 1.5B config (20 steps, 1.3 CFG) 2025-12-29 19:12:54 +08:00
Hawk Lee 521ceb67d7 Optimize 7B model quality with sampling parameters and update README 2025-12-29 19:10:05 +08:00
Hawk Lee f49c89d24c Enhance progress bar: jump to 100% on natural termination 2025-12-29 18:55:29 +08:00
Hawk Lee 3f2de07ecd Simplify VibeVoice TTS UI: remove max_length_times and set internal default 2025-12-29 18:53:32 +08:00
Hawk Lee 9263d3be70 Update README with official VibeVoice parameters and model-specific advice 2025-12-29 18:48:17 +08:00
Hawk Lee 7f65bfc749 Finalize VibeVoice integration: 7B support, resampling, punctuation fixes, and docs 2025-12-29 18:46:12 +08:00
Hawk Lee 520d7bdcc8 Fix 7B local path detection: use correct org name (vibevoice vs microsoft) 2025-12-29 18:37:39 +08:00
Hawk Lee b863cf74e5 Add normalize_text toggle to control text preprocessing 2025-12-29 18:36:27 +08:00
Hawk Lee 7f8f700e82 Note 1.5B punctuation limitations, recommend 7B for production 2025-12-29 18:34:37 +08:00
Hawk Lee 322ae81333 Make hyphen conversion more conservative: only year ranges with 年 2025-12-29 18:34:02 +08:00
Hawk Lee ac1237977c Add punctuation normalization: replace hyphens with '至', remove quotes 2025-12-29 18:32:29 +08:00
Hawk Lee 331cf7ac58 Document CFG scale speed effect and reference audio resampling 2025-12-29 18:21:43 +08:00
Hawk Lee d892c60f51 Add automatic resampling of reference audio to 24000Hz 2025-12-29 18:08:57 +08:00
Hawk Lee e02e02e0d2 Add debug prints to trace audio length issue 2025-12-29 18:05:52 +08:00
Hawk Lee 15d7a317ab Fix 7B download path to vibevoice/VibeVoice-7B 2025-12-29 17:54:07 +08:00
Hawk Lee ccfa97d03b Fix 7B path (vibevoice/VibeVoice-7B), add max_length_times param to prevent early cutoff 2025-12-29 17:52:12 +08:00
Hawk Lee 121bb6c61a Add speed control parameter (post-process time-stretch) 2025-12-29 17:47:17 +08:00
Hawk Lee 9f70ca0155 Add VibeVoice 7B support; update README with language info (en/zh only) 2025-12-29 17:40:44 +08:00
Hawk Lee 7f2dd14a1f Fix VibeVoice params: replace language/speed with cfg_scale/ddpm_steps 2025-12-29 17:35:02 +08:00
Hawk Lee abf78c0257 Update README: VibeVoice development status and known limitations 2025-12-29 17:30:47 +08:00
Hawk Lee c5b9317523 Final cleanup: remove debug prints, delete temp repos 2025-12-29 17:26:35 +08:00
Hawk Lee 3e7fee5939 Fix KeyError in non-streaming model: use resolved input_ids 2025-12-29 17:16:51 +08:00
Hawk Lee 8172878446 Fix config class mismatch: use non-streaming VibeVoiceConfig for non-streaming model 2025-12-29 17:12:17 +08:00
Hawk Lee 8ae16192de Remove temp_repo submodule 2025-12-29 17:07:48 +08:00
Hawk Lee be9f4b05f5 Switch to non-streaming VibeVoiceForConditionalGenerationInference 2025-12-29 17:07:32 +08:00
Hawk Lee 83072f76fe Pass input_ids positionally to fix transformers generation error 2025-12-29 16:53:33 +08:00
Hawk Lee e47d9c432e Refactor VibeVoice generate to handle inputs dynamically and set explicit token IDs 2025-12-29 16:38:04 +08:00
Hawk Lee a60810cdca Add debug logging for processor output keys 2025-12-29 16:17:52 +08:00
Hawk Lee d2022fc825 Bundle VibeVoice generation_config and implement fallback loading 2025-12-29 16:10:48 +08:00
Hawk Lee 848297a804 Fix UnboundLocalError in VibeVoice core 2025-12-29 15:48:09 +08:00
Hawk Lee d76d2712e6 Fix VibeVoice load path cleanup bug & update README file structure 2025-12-29 15:45:16 +08:00
Hawk Lee e8de3283b3 Refactor VibeVoice to self-contained core (no remote code execution) 2025-12-29 15:40:35 +08:00