Commit Graph

155 Commits

Author SHA1 Message Date
babysor00
c3590bffb2 Add description for 2022-07-17 11:56:13 +08:00
babysor00
efbdb21b70 Refactor model 2022-07-17 11:55:53 +08:00
Vega
6abdd0ebf0
Refactor (#649)
* Refactor model

* Refactor and fix bug to save plots
2022-07-17 09:58:17 +08:00
wenqingl
400a7207e3
Update README.md (#640) 2022-07-14 17:41:00 +08:00
babysor00
6f023e313d Add web gui of training and reconstruct taco model methods 2022-06-26 23:21:32 +08:00
babysor00
a39b6d3117 Remove breaking import for Macos 2022-06-26 11:56:50 +08:00
babysor00
885225045d 修复兼容性 - mac + linux 2022-06-25 20:17:06 +08:00
babysor00
ee643d7cbc Fix compatibility issue 2022-06-18 23:46:44 +08:00
flysmart
6a793cea84
Added missing files for Fre-GAN (#579)
* The new vocoder Fre-GAN is now supported

* Improved some fregan details

* Fixed the problem that the existing model could not be loaded to continue training when training GAN

* Updated reference papers

* GAN training now supports DistributedDataParallel (DDP)

* Added requirements.txt

* GAN training uses single card training by default

* Added note about GAN vocoder training with multiple GPUs

* Added missing files for Fre-GAN
2022-05-25 23:29:59 +08:00
Evers
7317ba5ffe
add gen_voice.py for handle by python command instead of demon_tool gui. (#560) 2022-05-22 16:28:58 +08:00
flysmart
05f886162c
GAN training now supports DistributedDataParallel (DDP) (#558)
* The new vocoder Fre-GAN is now supported

* Improved some fregan details

* Fixed the problem that the existing model could not be loaded to continue training when training GAN

* Updated reference papers

* GAN training now supports DistributedDataParallel (DDP)

* Added requirements.txt

* GAN training uses single card training by default

* Added note about GAN vocoder training with multiple GPUs
2022-05-22 16:24:50 +08:00
babysor00
e726c2eb12 Merge branch 'main' of https://github.com/babysor/Realtime-Voice-Clone-Chinese 2022-05-15 16:09:09 +08:00
babysor00
c00474525a Fix nits of file path 2022-05-15 16:08:58 +08:00
flysmart
350b190662
Solved the problem that the existing model could not be loaded when training the GAN model (#549)
* The new vocoder Fre-GAN is now supported

* Improved some fregan details

* Fixed the problem that the existing model could not be loaded to continue training when training GAN

* Updated reference papers
2022-05-13 13:41:03 +08:00
flysmart
0caed984e3
The new vocoder Fre-GAN is now supported (#546)
* The new vocoder Fre-GAN is now supported

* Improved some fregan details
2022-05-12 12:27:17 +08:00
Vega
c5d03fb3cb
Upgrade to new web service (#529)
* Init new GUI

* Remove unused codes

* Reset layout

* Add samples

* Make framework to support multiple pages

* Add vc mode

* Add preprocessing mode

* Add training mode

* Remove text input in vc mode

* Add entry for GUI and revise readme

* Move requirement together

* Add error raise when no model folder found

* Add readme
2022-05-09 18:44:02 +08:00
babysor00
7f799d322f Tell the hifigan type of a vocoder model by searching full text 2022-04-30 10:31:01 +08:00
LZY
a1f2e4a790
Update README-CN.md (#523) 2022-04-28 16:16:37 +08:00
Lix Zhou
b136f80f43
Add an aliyunpan download link (#505)
Baidu Yun Pan is so fxxking slow
2022-04-26 21:20:39 +08:00
Moose W. Oler
f082a82420
fix issue #496 (#498)
pass `wav`, `sampling_rate` (in encoder/audio.py line 59 ) as keyword args instead of postional args to prevent warning messages from massing up console outputs while adopting librosa 0.9.1 occasionally.
2022-04-11 17:26:52 +08:00
babysor00
7f0d983da7 Merge branch 'main' of https://github.com/babysor/Realtime-Voice-Clone-Chinese 2022-04-02 18:11:52 +08:00
babysor00
0353bfc6e6 New GUI in order to combine web and toolbox in future 2022-04-02 18:11:49 +08:00
Vega
9ec114a7c1
Create FUNDING.yml 2022-04-02 10:16:02 +08:00
1itt1eB0y
ddf612e87c
Update README-CN.md (#470)
修正一个简单的翻译问题
2022-03-24 12:52:47 +08:00
babysor00
374cc89cfa Fix web generate with rnn bug 2022-03-19 12:16:55 +08:00
babysor00
6009da7072 Merge branch 'main' of https://github.com/babysor/Realtime-Voice-Clone-Chinese 2022-03-19 12:14:24 +08:00
babysor00
1c61a601d1 Remove dependency of pyworld for non-vc mode 2022-03-19 12:14:21 +08:00
Vega
02ee514aa3
Update issue templates 2022-03-12 19:28:03 +08:00
babysor00
6532d65153 Add default param for inferencing of vocoder 2022-03-12 19:14:01 +08:00
babysor00
3fe0690cc6 Update README-CN.md 2022-03-09 09:39:24 +08:00
babysor00
79f424d614 Add default path for hifi 2022-03-08 09:17:56 +08:00
babysor00
3c97d22938 Fix bug when searching for vocoder 2022-03-07 20:08:29 +08:00
babysor00
fc26c38152 Fix compatibility issue 2022-03-07 17:05:22 +08:00
babysor00
6c01b92703 Fix bug introduced by config file reading 2022-03-06 09:35:25 +08:00
babysor00
c36f02634a Add link and separate requirement file 2022-03-05 12:20:09 +08:00
Vega
b05e7441ff
Fix nit in readme 2022-03-05 00:55:08 +08:00
Vega
693de98f4d
Add instruction image 2022-03-05 00:54:31 +08:00
Vega
252a5e11b3
Ppg vc init (#421)
* Init  ppg extractor and ppg2mel

* add preprocess and training

* FIx known issues

* Update __init__.py

Allow to gen audio

* Fix length issue

* Fix bug of preparing fid

* Fix sample issues

* Add UI usage of PPG-vc

* Add readme
2022-03-05 00:52:36 +08:00
Vega
b617a87ee4
Init ppg extractor and ppg2mel (#375)
* Init  ppg extractor and ppg2mel

* add preprocess and training

* FIx known issues

* Update __init__.py

Allow to gen audio

* Fix length issue

* Fix bug of preparing fid

* Fix sample issues

* Add UI usage of PPG-vc
2022-03-03 23:38:12 +08:00
AyahaShirane
ad22997614
fixed the issues #372 (#379)
修复了一些参数传递造成的问题,把过时的torch.nn.functional.tanh()改成了torch.tanh()
2022-02-27 11:02:01 +08:00
hertz
9e072c2619
Hifigan Support train from existed checkpoint. (#389)
* 1k steps to save tmp hifigan model

* hifigan support train from existed ckpt
2022-02-27 11:01:47 +08:00
Alex Newton
b79e9d68e4
连续换行造成的多了个None (#405)
小问题,gui好像没有这个问题,自己测试web的时候直接调用的函数发现的这个情况
2022-02-27 10:55:00 +08:00
babysor00
0536874dec Add config file for pretrained 2022-02-23 09:37:39 +08:00
李子
4529479091
指定librosa版本 (#378)
* 支持data_aishell(SLR33)数据集

* 更新readme

* 指定librosa版本
2022-02-10 20:47:26 +08:00
babysor00
8ad9ba2b60 change naming logic of saving trained file for synthesizer to allow shorter interval 2022-01-15 17:56:14 +08:00
D-Blue
b56ec5ee1b
Fix a UserWarning (#273)
Fix a UserWarning in synthesizer/synthesizer_dataset.py, because of converting list of numpy array to torch tensor at Ln.85.
2021-12-20 20:33:12 +08:00
CrystalRays
0bc34a5bc9
Fix TypeError at line 459 in toolbox/ui.py when both PySide6(PyQt6) and PyQt5 installed (#255)
### Error Info Screenshot
![](https://cdn.jsdelivr.net/gh/CrystalRays/CDN@main/img/16389623959301638962395845.png)

### Error Reason
Matplotlib.backends.qt_compat.py decide the version of qt library according to sys.modules firstly, os.environ secondly and the sequence of PyQt6, PySide6, PyQt5, PySide 2 and etc finally. Import PyQt5 after matplotlib make that there is no PyQt5 in sys.modules so that it choose PyQt6 or PySide6 before PyQt5 if it installed.
因为Matplotlib.backends.qt_compat.py优先根据导入的库决定要使用的Python Qt的库,如果没有导入则根据环境变量PYQT_APT决定,再不济就按照PyQt6, PySide6, PyQt5, PySide 2的顺序导入已经安装的库。因为ui.py先导入matplotlib而不是PYQT5导致matplotlib在导入的库里找不到Qt的库,又没有指定环境变量,然后用户安装了Qt6的库的话就导入Qt6的库去了
2021-12-15 12:41:10 +08:00
Wings Music
875fe15069
Update readme for training encoder (#250) 2021-12-07 19:10:29 +08:00
zzxiang
4728863f9d
Fix inference on cpu device (#241) 2021-11-29 21:10:07 +08:00
hertz
a4daf42868
1k steps to save tmp hifigan model (#240) 2021-11-29 21:09:54 +08:00