blackteay/GPT-SoVITS
1
0
Fork 0
mirror of https://github.com/RVC-Boss/GPT-SoVITS.git synced 2025-12-18 02:47:02 +08:00
Code Issues Packages Projects Releases Wiki Activity
1,065 Commits 3 Branches 4 Tags
Commit Graph

425 Commits

Author SHA1 Message Date
zpeng11
142414430b
Merge 8858492f56326dce521db2c4a4b3a7323e786596 into 36b3231c6f7b1adabb1e93bb3c6449dbd85b7375 2025-12-15 14:37:12 +08:00
ChasonJiang
36b3231c6f
bug fix (#2689) 2025-12-15 14:23:06 +08:00
RVC-Boss
cb00840c4e
Add files via upload 2025-11-28 22:02:03 +08:00
wzy3650
60a4a214af
vq distributed training support (#2577) 
Co-authored-by: wangzeyuan <wangzeyuan@agora.io>
 2025-11-28 21:57:13 +08:00
zzz
6375bbe316
尝试 stream infer (#2469) 
* 尝试 stream infer

* 在 stream_infer 脚本中绘制生成的音频

* stream_infer 增加导出部分。

* stream_infer: 更方便找规律的图

* stream_infer: 在拼接音频时进行相关性搜索，减少拼接带来基频断裂的情况

* stream_infer: 导出 `find_best_audio_offset_fast`

* stream_infer: 优化波形显示，方便对比

* stream_v2pro.py 从命令行读取参数

* stream_v2pro.py 减少用于导出的文本长度

* stream_v2pro: 修复由于 spectrogram_torch 输入是 half 导致 spec 溢出最终没有声音的问题

* stream_v2pro: 新增 --lang 参数提示参考文字的语言类型
 2025-11-28 21:36:57 +08:00
KamioRinn
e00ca92140
Fix ASMD (#2636) 2025-11-28 21:22:43 +08:00
ChasonJiang
92ab59c553
更细粒度的流式推理模式 (#2671) 
* 更好的流式推理模式

* 清理无用代码

* modified:   GPT_SoVITS/AR/models/t2s_model.py
	modified:   GPT_SoVITS/TTS_infer_pack/TTS.py
	modified:   GPT_SoVITS/module/models.py

* modified:   GPT_SoVITS/TTS_infer_pack/TTS.py

* modified:   .gitignore
	modified:   GPT_SoVITS/AR/models/t2s_model.py
	modified:   GPT_SoVITS/TTS_infer_pack/TTS.py
	modified:   GPT_SoVITS/module/models.py

* modified:   GPT_SoVITS/AR/models/t2s_model.py
	modified:   GPT_SoVITS/TTS_infer_pack/TTS.py
	modified:   GPT_SoVITS/module/models.py
	modified:   api_v2.py

* modified:   GPT_SoVITS/TTS_infer_pack/TTS.py

* 更正拼写错误

* 支持固定chunk长度的流式推理，优化sola算法

* 修复api_v2的ogg格式传输问题
 2025-11-28 21:12:41 +08:00
zpeng11
8858492f56 feat:enable speed control for v1v2 2025-08-31 00:14:07 -04:00
zpeng11
337da7454e feat:remove prints 2025-08-26 17:09:36 -04:00
zpeng11
3e63595f0e feat:update kv cache to [len, head, dim] to allow linear size increasement 2025-08-26 17:03:29 -04:00
zpeng11
fa84e262ae feat:remove unneed for main 2025-08-25 22:53:15 -04:00
zpeng11
968ac4c264 feat: solved problem, export works 2025-08-25 22:37:52 -04:00
zpeng11
419909b443 failed , testing expand y 2025-08-25 21:57:36 -04:00
zpeng11
c85ee3d521 feat:successfully unified first step and following step 2025-08-25 17:57:04 -04:00
zpeng11
d413a4f5b1 run time working 2025-08-25 17:07:38 -04:00
zpeng11
26228402e3 feat:solve unified kv cache shape handling, todo: clean up upper level to unify first and following step 2025-08-25 12:06:26 -04:00
zpeng11
0c5f61f98c feat:rename and features to onnx export 2025-08-25 01:46:53 -04:00
zpeng11
942caa888e feat:supporting half export 2025-08-24 02:29:33 -04:00
zpeng11
48d52778ce feat:clean up export logics and add notes 2025-08-24 02:00:32 -04:00
zpeng11
e4d1894a8f feat:experiments with for onnx with attention, but does not work well todo:clean code and try v3v4 2025-08-24 00:46:29 -04:00
zpeng11
5982080939 feat:updated fsdecode and decoder interface 2025-08-23 17:35:21 -04:00
zpeng11
b45cbc3561 feat: sampling params working now for export, todo:fold weights clean code 2025-08-23 13:03:02 -04:00
zpeng11
9ed42daa88 feat: allow fsdec and sdec to have sampling parames 2025-08-23 12:17:04 -04:00
zpeng11
3ccd1c0ea3 fix: solved t2s ending problem, and verified infer&forward has same output under deterministic random, fixed topk to 15 2025-08-23 03:31:01 -04:00
zpeng11
63cbb6efa7 verified audio, text, synthesizer all working, todo:dig into t2s for error 2025-08-22 01:36:58 -04:00
zpeng11
e8fdf472c0 feat:onnx friendly loop with same function 2025-08-21 22:23:50 -04:00
zpeng11
77794a5923 feat:export onnx with combined graph ready, todo:link weights in onnx graph 2025-08-21 01:52:34 -04:00
zpeng11
16d30ce1e4 feat:get ready for if node merge 2025-08-21 00:34:56 -04:00
zpeng11
403c5bf320 feat:v1v2 both works for export 2025-08-20 20:57:29 -04:00
zpeng11
bc7fe01876 feat:update v2pro's gpt path, todo:work on v1 transform 2025-08-20 20:41:20 -04:00
zpeng11
bb529e7e4a update namings 2025-08-20 20:21:42 -04:00
zpeng11
4e0cc57052 update init_step name 2025-08-20 20:05:07 -04:00
zpeng11
aafa0561d8 correctly setup onnx export, solved problem 2025-08-20 19:39:10 -04:00
zpeng11
94b31a250f limit vits for one input a time 2025-08-20 18:39:21 -04:00
zpeng11
fd0fb35a49 fix spectrum take out working 2025-08-20 18:32:38 -04:00
zpeng11
911c53b1ee fixed using hubert for full run, 80 works 2025-08-20 17:37:41 -04:00
zpeng11
1cdd41d877 fixed resample audio and tested in full run 2025-08-20 16:47:55 -04:00
zpeng11
da5aa78224 feat:combined fsdc and encoder, todo:extract audio pipeline 2025-08-20 02:24:59 -04:00
zpeng11
5c08328cf3 feat:voice and text preprocess system verifed, todo:dissasemble onnx export of gsv 2025-08-19 21:10:21 -04:00
zpeng11
4e42a28f9c feat: v2pro and v2proplus onnx export works, todo:solve data dependencies 2025-08-18 20:19:27 -04:00
zpeng11
610b36561a feat:remove debug, todo:rewrite the onnx export interface 2025-08-17 19:22:11 -04:00
zpeng11
8c0f32da3e feat:v2pp onnx export ready testing... 2025-08-17 17:54:57 -04:00
ChasonJiang
b5a67e6247
修复gpt的loss计算问题 (#2537) 
* 修复gpt的loss计算问题

* fallback tts config
 2025-07-18 14:59:59 +08:00
ChasonJiang
b9211657d8
优化TTS_Config的代码逻辑 (#2536) 
* 优化TTS_Config的代码逻辑

* 在载入vits权重之后保存tts_config
 2025-07-18 11:54:40 +08:00
RVC-Boss
2d09bbe63a
Update tts_infer.yaml 2025-07-16 15:44:04 +08:00
RVC-Boss
4d8ebf8523
Update TTS.py 2025-07-16 15:43:26 +08:00
jiangsier-xyz
e476b01f30
解决 TTS.py 无法识别真正支持版本 v2Pro、v2ProPlus 的问题 (#2490) 
同时更新一版默认配置。

Co-authored-by: jiangsier-xyz <jiangsier131@gmail.com>
 2025-07-16 15:42:36 +08:00
RVC-Boss
426e1a2bb4
提升推理进程优先级 2025-07-10 18:16:45 +08:00
RVC-Boss
4e3c69043c
Update inference_webui.py 2025-07-10 18:16:24 +08:00
Yixiao Chen
8c579d46dd
Update export_torch_script.py (#2494) 
Avoid dtype inconsistency when exporting
 2025-07-02 22:48:28 +08:00