 Hawk Lee
|
5ea20cb94f
|
Update README
|
2026-01-04 18:17:57 +08:00 |
|
|
Hawk Lee
|
68d5df99d8
|
Fix: Remove explicit resampling logic from VoxCPM node (model is now native 44.1k)
|
2026-01-04 18:00:14 +08:00 |
|
|
Hawk Lee
|
f9bff68f07
|
Fix: Hardcode AudioVAE default params to 44.1kHz (VoxCPM 1.5 Native) to bypass config loading issues
|
2026-01-04 17:42:31 +08:00 |
|
|
Hawk Lee
|
ba9a62267c
|
Refactor: Fully replace vendored VoxCPM core with Wildminder's High-Fidelity implementation (src/voxcpm)
|
2026-01-04 17:32:17 +08:00 |
|
|
Hawk Lee
|
2e0a00f56b
|
Feat: Resample output to 44.1kHz to meet user expectations (upsampling 16k->44.1k)
|
2026-01-04 17:09:32 +08:00 |
|
|
Hawk Lee
|
de6b1084dc
|
Revert: Restore 44.1kHz sample rate as user confirmed file spectrum is correct (fast speed is pace, not pitch)
|
2026-01-04 17:04:59 +08:00 |
|
|
Hawk Lee
|
ef3a9b6320
|
Fix: Force 16kHz output rate to resolve fast playback speed (despite 44.1k config claim)
|
2026-01-04 16:58:33 +08:00 |
|
|
Hawk Lee
|
e2b235ff43
|
Fix: correct sample rate from model metadata (fixes fast playback) and remove debug prints
|
2026-01-04 16:44:53 +08:00 |
|
|
Hawk Lee
|
edeb239efa
|
Debug: Add print statements to trace audio tensor shape issues
|
2026-01-04 16:39:41 +08:00 |
|
|
Hawk Lee
|
a25ba3946c
|
Fix: Output 3D audio tensor [Batch, Channel, Time] for ComfyUI compatibility
|
2026-01-04 16:37:04 +08:00 |
|
|
Hawk Lee
|
95c60998b6
|
Fix: Correct indentation of try-except block in generate method
|
2026-01-04 16:33:38 +08:00 |
|
|
Hawk Lee
|
2bb1633e0f
|
Fix: Indentation error in generate method
|
2026-01-04 16:30:33 +08:00 |
|
|
Hawk Lee
|
a79d308030
|
Fix: Restore missing try-block causing SyntaxError
|
2026-01-04 16:28:56 +08:00 |
|
|
Hawk Lee
|
0f214d10bf
|
Fix: Ensure prompt_text is None if no audio prompt is provided (fixes VoxCPM strict validation)
|
2026-01-04 16:27:00 +08:00 |
|
|
Hawk Lee
|
6d2483a983
|
Fix: Align VoxCPM generate arguments (rename prompt_wav_path, add prompt_text, remove unsupported params)
|
2026-01-04 16:23:05 +08:00 |
|
|
Hawk Lee
|
e3a6b4c721
|
Fix: Disable torch.compile (optimize=False) to resolve cudaMallocAsync crash
|
2026-01-04 16:20:53 +08:00 |
|
|
Hawk Lee
|
6ae43bc55d
|
Fix(UI): Remove long labels from enable_denoiser to prevent text overlap
|
2026-01-04 16:19:09 +08:00 |
|
|
Hawk Lee
|
7b258fee31
|
Docs: Update ZipEnhancer path to local models dir
|
2026-01-04 16:15:50 +08:00 |
|
|
Hawk Lee
|
2fdb460f26
|
Fix: Download ZipEnhancer to local models dir instead of ~/.cache
|
2026-01-04 16:15:43 +08:00 |
|
|
Hawk Lee
|
6b50b1744c
|
Docs: Add ZipEnhancer manual download instructions
|
2026-01-04 16:14:18 +08:00 |
|
|
Hawk Lee
|
de20148036
|
Feat: Add 'enable_denoiser' toggle to VoxCPM Loader
|
2026-01-04 16:13:10 +08:00 |
|
|
Hawk Lee
|
e04e0b907e
|
Fix: Restore base_path definition for locating core module
|
2026-01-04 16:01:32 +08:00 |
|
|
Hawk Lee
|
ad766435d2
|
Fix: Remove 'device' arg from VoxCPM init (not supported by core)
|
2026-01-04 15:58:37 +08:00 |
|
|
Hawk Lee
|
5d122b74c0
|
Refactor: Vendor VoxCPM core code to avoid pip dependency hell
|
2026-01-04 15:55:59 +08:00 |
|
|
Hawk Lee
|
b4ac7603dd
|
Fix: Remove auto-install to prevent dependency hell (VoxCPM requires Torch>=2.5)
|
2026-01-04 15:51:52 +08:00 |
|
|
Hawk Lee
|
bd3ba6caa8
|
Feat: Auto-install voxcpm dependency if missing
|
2026-01-04 15:50:09 +08:00 |
|
|
Hawk Lee
|
ca6181a6df
|
Fix: Switch to 'voxcpm' python package for loading
|
2026-01-04 15:49:03 +08:00 |
|
|
Hawk Lee
|
2b66f12ac2
|
Fix: Move VoxCPM model path to global ComfyUI/models dir
|
2026-01-04 15:46:22 +08:00 |
|
|
Hawk Lee
|
6ab48f0adc
|
Feat: Implement VoxCPM Inference Logic
|
2026-01-04 15:43:24 +08:00 |
|
|
Hawk Lee
|
392fb5dd9c
|
Docs: Fix VoxCPM Repo ID in README
|
2026-01-04 15:38:41 +08:00 |
|
|
Hawk Lee
|
5c979556a9
|
Fix: Update VoxCPM Repo ID to openbmb/VoxCPM1.5
|
2026-01-04 15:38:34 +08:00 |
|
|
Hawk Lee
|
8f1cde850e
|
Docs: Add VoxCPM 1.5 manual download instructions
|
2026-01-04 15:35:36 +08:00 |
|
|
Hawk Lee
|
d0d652446b
|
Feat: Add VoxCPM 1.5 Node Skeleton
|
2026-01-04 15:34:42 +08:00 |
|
|
Hawk Lee
|
6cfa1274e2
|
Docs: Add Voice Conversion support matrix (CosyVoice wins Audio-2-Audio)
|
2026-01-04 15:27:22 +08:00 |
|
|
Hawk Lee
|
00ee3c0b35
|
Docs: Add VoxCPM/CosyVoice/VibeVoice detailed comparison table
|
2026-01-04 15:20:30 +08:00 |
|
|
Hawk Lee
|
3147e5e997
|
Docs: Add context length comparison (Unlimited Streaming) to Sec 3.5
|
2026-01-04 15:12:02 +08:00 |
|
|
Hawk Lee
|
185bab0507
|
Release: v1.5.0 - Unified CosyVoice Fix & Updated Docs
|
2026-01-04 15:03:53 +08:00 |
|
|
Hawk Lee
|
9c0228af5c
|
Docs: Correct language recommendation to English only
|
2026-01-04 15:00:07 +08:00 |
|
|
Hawk Lee
|
c26ca18f20
|
Docs: Clarify language support for 300M-Instruct
|
2026-01-04 14:56:27 +08:00 |
|
|
Hawk Lee
|
55afa827ea
|
Fix: 300M-Instruct instruction following via <|endofprompt|> and update README
|
2026-01-04 14:50:26 +08:00 |
|
|
Hawk Lee
|
0913266ea2
|
Fix: Protect <|endofprompt|> from normalization in V1 Instruct
|
2026-01-04 12:54:08 +08:00 |
|
|
Hawk Lee
|
bad3b04619
|
Fix: Final restoration of 300M-Instruct (Native Surgical Path + Correct Boundary Tokens)
|
2026-01-04 00:30:31 +08:00 |
|
|
Hawk Lee
|
fff0f1bd11
|
Fix: Restored V1 instruction assembly syntax and switched to native inference_instruct
|
2026-01-04 00:23:26 +08:00 |
|
|
Hawk Lee
|
b3a6f4ccf2
|
Fix: Corrected spk2info key and refined V1 Instruct identity injection
|
2026-01-03 23:58:31 +08:00 |
|
|
Hawk Lee
|
56f51b4217
|
Fix: Final restoration for 300M-Instruct (boundary tokens and identity preservation)
|
2026-01-03 23:51:22 +08:00 |
|
|
Hawk Lee
|
22e2d9784a
|
Fix: Ensure 300M-Instruct uses the surgical instruction path in identity-only mode
|
2026-01-03 23:40:08 +08:00 |
|
|
Hawk Lee
|
c9df64d283
|
Assets: Added high-quality female and male seeds for CosyVoice V2/V3
|
2026-01-03 23:25:32 +08:00 |
|
|
Hawk Lee
|
fb68fa6421
|
Optimize: V3 seed quality and V1 Instruct following
|
2026-01-03 20:01:50 +08:00 |
|
|
Hawk Lee
|
584d8e234d
|
Baseline: Unified CosyVoice V1/V2/V3 support and stability fixes
|
2026-01-03 19:36:07 +08:00 |
|
|
Hawk Lee
|
b6b3b09b57
|
fix: restore 100% native paths for 300M models and remove hacks
|
2026-01-01 16:19:58 +08:00 |
|