Hugging Face's logo

MondMeer
/
GPT-SoVITS-HiiragiHakua

Text-to-Speech
Chinese
GPT-SoVITS
柊白亜
Hakua
Model card Files Community
MondMeer commited on
Commit
f779c16
·
1 Parent(s): c106cc2

l and l_nh weights

Browse files
Files changed (16) hide show
  1. GPT_weights_v2/{hakua1-e10.ckpt → hakua_f-e10.ckpt} +0 -0
  2. GPT_weights_v2/{hakua1-e15.ckpt → hakua_f-e15.ckpt} +0 -0
  3. GPT_weights_v2/{hakua1-e5.ckpt → hakua_f-e5.ckpt} +0 -0
  4. GPT_weights_v2/hakua_l-e10.ckpt +3 -0
  5. GPT_weights_v2/hakua_l-e15.ckpt +3 -0
  6. GPT_weights_v2/hakua_l-e5.ckpt +3 -0
  7. GPT_weights_v2/hakua_l_nh-e10.ckpt +3 -0
  8. GPT_weights_v2/hakua_l_nh-e15.ckpt +3 -0
  9. GPT_weights_v2/hakua_l_nh-e5.ckpt +3 -0
  10. README.md +21 -8
  11. SoVITS_weights_v2/{hakua1_e4_s1192.pth → hakua_f_e4_s1192.pth} +0 -0
  12. SoVITS_weights_v2/{hakua1_e8_s2384.pth → hakua_f_e8_s2384.pth} +0 -0
  13. SoVITS_weights_v2/hakua_l_e4_s336.pth +3 -0
  14. SoVITS_weights_v2/hakua_l_e8_s672.pth +3 -0
  15. SoVITS_weights_v2/hakua_l_nh_e4_s200.pth +3 -0
  16. SoVITS_weights_v2/hakua_l_nh_e8_s400.pth +3 -0
GPT_weights_v2/{hakua1-e10.ckpt → hakua_f-e10.ckpt} RENAMED
File without changes
GPT_weights_v2/{hakua1-e15.ckpt → hakua_f-e15.ckpt} RENAMED
File without changes
GPT_weights_v2/{hakua1-e5.ckpt → hakua_f-e5.ckpt} RENAMED
File without changes
GPT_weights_v2/hakua_l-e10.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c4bba74f8da4b4c237cdc4e7c9bf516b2582eac02fb6ee8ea608bb80ad4b13d
3
+ size 155312658
GPT_weights_v2/hakua_l-e15.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:54bac1533d881403fe8ba450261e2f3e490e55624f9d1ecfa6ee58f38cf19edf
3
+ size 155312957
GPT_weights_v2/hakua_l-e5.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3e6fbea92c2ac2dbac320262dceeac1e032e86bf28b1270e2597f29327e04dc
3
+ size 155312658
GPT_weights_v2/hakua_l_nh-e10.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa3fab2b551d474372064d11e52ec1217d1c9734ed9ea97ef02f02f65df60564
3
+ size 155312502
GPT_weights_v2/hakua_l_nh-e15.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:caf21a192fb276825a0e75c679c9324a9bcb83704af6d5691281d2146415e423
3
+ size 155312502
GPT_weights_v2/hakua_l_nh-e5.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:73fd123360bc2636ef244456b2e9e0bc0a3ce437f8c894e20c6a98ed8ce63841
3
+ size 155312502
README.md CHANGED
@@ -13,23 +13,36 @@ tags:
13
 
14
  ## 模型介绍
15
 
16
- * 基于 [GPT-SoVITS](https://github.com/RVC-Boss/GPT-SoVITS) 项目训练的 [柊 白亞](https://zh.moegirl.org.cn/zh-hans/柊白亚) 声音模型
17
- * 使用游戏内约3000条语音微调
18
- * 推理时建议按句号切。
19
 
20
- 示例：
21
 
22
  <audio controls>
23
  <source src="examples\hakua.wav" type="audio/wav">
24
- Your browser does not support the audio element.
25
  </audio>
26
 
27
  <audio controls>
28
  <source src="examples\0721.wav" type="audio/mpeg">
29
- Your browser does not support the audio element.
30
  </audio>
31
 
32
  <audio controls>
33
  <source src="examples\yinhetiedao.wav" type="audio/mpeg">
34
- Your browser does not support the audio element.
35
- </audio>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
 
14
  ## 模型介绍
15
 
16
+ * 基于 [GPT-SoVITS](https://github.com/RVC-Boss/GPT-SoVITS) 项目训练的 [柊白亞](https://zh.moegirl.org.cn/zh-hans/柊白亚) 声音模型
17
+ * 推理时建议按句号切。句子太长可以用标点符号切。
 
18
 
19
+ 示例（使用hakua_f_e8_s2384和hakua_f-e15。实际上使用hakua_l的效果明显更好）：
20
 
21
  <audio controls>
22
  <source src="examples\hakua.wav" type="audio/wav">
 
23
  </audio>
24
 
25
  <audio controls>
26
  <source src="examples\0721.wav" type="audio/mpeg">
 
27
  </audio>
28
 
29
  <audio controls>
30
  <source src="examples\yinhetiedao.wav" type="audio/mpeg">
31
+ </audio>
32
+
33
+ ## 模型列表
34
+
35
+ | 模型名称（前缀） | 数据集 |
36
+ | :--------------- | :----------------------------------------------------------------------------------------------------- |
37
+ | hakua_f | 使用游戏内几乎所有语音(大概一百条语音未被匹配，没有仔细找原因) 共 2934 条语音记录 总语音时长: 04:31:18 |
38
+ | hakua_l | 使用游戏内大于等于五秒的语音 共 1537 条语音记录 总语音时长: 03:19:59 |
39
+ | hakua_l_nh | 使用游戏内大于等于五秒且非h语音 共 934 条语音记录 总语音时长: 01:53:32 |
40
+
41
+ 精简后的数据集，hakua_l以及hakua_l_nh的效果明显更好。
42
+
43
+ nh版本与l版本区别似乎并不大。
44
+
45
+ ## TODO
46
+
47
+ - [ ] 挑选合适的参考音频
48
+ - [ ] 训练RVC模型
SoVITS_weights_v2/{hakua1_e4_s1192.pth → hakua_f_e4_s1192.pth} RENAMED
File without changes
SoVITS_weights_v2/{hakua1_e8_s2384.pth → hakua_f_e8_s2384.pth} RENAMED
File without changes
SoVITS_weights_v2/hakua_l_e4_s336.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cef92e102f3e444662fe2a6bac3824ddb0e8a9b244b5390ae8938c8213fa9e26
3
+ size 85005848
SoVITS_weights_v2/hakua_l_e8_s672.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d7b74c3128f99829b5ec5c37854e3d02b92c46ad65d9126476b54d9332486c7
3
+ size 85007879
SoVITS_weights_v2/hakua_l_nh_e4_s200.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13e72c0507dea13ead8994fc9a1d0ccd034ce080df4d93b8e193b3f6953314e8
3
+ size 85007424
SoVITS_weights_v2/hakua_l_nh_e8_s400.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:53ac63e8c4be4265d6e71893877cb96fdb3c24abae1849486d1a1edd48f4cc3b
3
+ size 85007424