mirror of
https://github.com/RVC-Boss/GPT-SoVITS.git
synced 2025-08-22 19:19:47 +08:00
docs(中文更新日志): 重新排版24年08月以来的更新日志
This commit is contained in:
parent
968952fd2a
commit
9903f66bc0
@ -1,4 +1,6 @@
|
|||||||
### 20240121更新
|
# 更新日志
|
||||||
|
|
||||||
|
## 20240121
|
||||||
|
|
||||||
1-config添加is_share, 诸如colab等场景可以将此改为True, 来使得webui映射到公网
|
1-config添加is_share, 诸如colab等场景可以将此改为True, 来使得webui映射到公网
|
||||||
|
|
||||||
@ -12,7 +14,7 @@
|
|||||||
|
|
||||||
6-大幅削弱合成音频包含参考音频结尾的问题
|
6-大幅削弱合成音频包含参考音频结尾的问题
|
||||||
|
|
||||||
### 20240122更新
|
## 20240122
|
||||||
|
|
||||||
1-修复过短输出文件返回重复参考音频的问题.
|
1-修复过短输出文件返回重复参考音频的问题.
|
||||||
|
|
||||||
@ -20,7 +22,7 @@
|
|||||||
|
|
||||||
3-音频路径检查.如果尝试读取输入错的路径报错路径不存在, 而非ffmpeg错误.
|
3-音频路径检查.如果尝试读取输入错的路径报错路径不存在, 而非ffmpeg错误.
|
||||||
|
|
||||||
### 20240123更新
|
## 20240123
|
||||||
|
|
||||||
1-解决hubert提取nan导致SoVITS/GPT训练报错ZeroDivisionError的问题
|
1-解决hubert提取nan导致SoVITS/GPT训练报错ZeroDivisionError的问题
|
||||||
|
|
||||||
@ -30,7 +32,7 @@
|
|||||||
|
|
||||||
4-中文分词使用jieba_fast代替jieba
|
4-中文分词使用jieba_fast代替jieba
|
||||||
|
|
||||||
### 20240126更新
|
## 20240126
|
||||||
|
|
||||||
1-支持输出文本中英混合、日英混合
|
1-支持输出文本中英混合、日英混合
|
||||||
|
|
||||||
@ -46,7 +48,7 @@
|
|||||||
|
|
||||||
7-自动识别不支持半精度的卡强制单精度.cpu推理下强制单精度.
|
7-自动识别不支持半精度的卡强制单精度.cpu推理下强制单精度.
|
||||||
|
|
||||||
### 20240128更新
|
## 20240128
|
||||||
|
|
||||||
1-修复数字转汉字念法问题
|
1-修复数字转汉字念法问题
|
||||||
|
|
||||||
@ -58,7 +60,7 @@
|
|||||||
|
|
||||||
5-完善Dockerfile的下载模型流程
|
5-完善Dockerfile的下载模型流程
|
||||||
|
|
||||||
### 20240129更新
|
## 20240129
|
||||||
|
|
||||||
1-16系等半精度训练有问题的显卡把训练配置改为单精度训练
|
1-16系等半精度训练有问题的显卡把训练配置改为单精度训练
|
||||||
|
|
||||||
@ -67,7 +69,7 @@
|
|||||||
3-修复git clone modelscope funasr仓库+老版本funasr导致接口不对齐报错的问题
|
3-修复git clone modelscope funasr仓库+老版本funasr导致接口不对齐报错的问题
|
||||||
|
|
||||||
|
|
||||||
### 20240130更新
|
## 20240130
|
||||||
|
|
||||||
1-所有涉及路径的地方双引号自动去除,小白复制路径带双引号不会报错
|
1-所有涉及路径的地方双引号自动去除,小白复制路径带双引号不会报错
|
||||||
|
|
||||||
@ -75,19 +77,19 @@
|
|||||||
|
|
||||||
3-增加按标点符号切分
|
3-增加按标点符号切分
|
||||||
|
|
||||||
### 20240201更新
|
## 20240201
|
||||||
|
|
||||||
1-修复uvr5读取格式错误导致分离失败的问题
|
1-修复uvr5读取格式错误导致分离失败的问题
|
||||||
|
|
||||||
2-支持中日英混合多种文本自动切分识别语种
|
2-支持中日英混合多种文本自动切分识别语种
|
||||||
|
|
||||||
### 20240202更新
|
## 20240202
|
||||||
|
|
||||||
1-修复asr路径尾缀带/保存文件名报错
|
1-修复asr路径尾缀带/保存文件名报错
|
||||||
|
|
||||||
2-引入paddlespeech的Normalizer https://github.com/RVC-Boss/GPT-SoVITS/pull/377 修复一些问题, 例如: xx.xx%(带百分号类), 元/吨 会读成 元吨 而不是元每吨,下划线不再会报错
|
2-引入paddlespeech的Normalizer https://github.com/RVC-Boss/GPT-SoVITS/pull/377 修复一些问题, 例如: xx.xx%(带百分号类), 元/吨 会读成 元吨 而不是元每吨,下划线不再会报错
|
||||||
|
|
||||||
### 20240207更新
|
## 20240207
|
||||||
|
|
||||||
1-修正语种传参混乱导致中文推理效果下降 https://github.com/RVC-Boss/GPT-SoVITS/issues/391
|
1-修正语种传参混乱导致中文推理效果下降 https://github.com/RVC-Boss/GPT-SoVITS/issues/391
|
||||||
|
|
||||||
@ -103,29 +105,29 @@
|
|||||||
|
|
||||||
7-集成faster whisper ASR日文英文
|
7-集成faster whisper ASR日文英文
|
||||||
|
|
||||||
### 20240208更新
|
## 20240208
|
||||||
|
|
||||||
1-GPT训练卡死 (win10 1909) 和https://github.com/RVC-Boss/GPT-SoVITS/issues/232 (系统语言繁体) GPT训练报错, [尝试修复](https://github.com/RVC-Boss/GPT-SoVITS/commit/59f35adad85815df27e9c6b33d420f5ebfd8376b).
|
1-GPT训练卡死 (win10 1909) 和https://github.com/RVC-Boss/GPT-SoVITS/issues/232 (系统语言繁体) GPT训练报错, [尝试修复](https://github.com/RVC-Boss/GPT-SoVITS/commit/59f35adad85815df27e9c6b33d420f5ebfd8376b).
|
||||||
|
|
||||||
### 20240212更新
|
## 20240212
|
||||||
|
|
||||||
1-faster whisper和funasr逻辑优化.faster whisper转镜像站下载, 规避huggingface连不上的问题.
|
1-faster whisper和funasr逻辑优化.faster whisper转镜像站下载, 规避huggingface连不上的问题.
|
||||||
|
|
||||||
2-DPO Loss实验性训练选项开启, 通过构造负样本训练缓解GPT重复漏字问题.推理界面公开几个推理参数. https://github.com/RVC-Boss/GPT-SoVITS/pull/457
|
2-DPO Loss实验性训练选项开启, 通过构造负样本训练缓解GPT重复漏字问题.推理界面公开几个推理参数. https://github.com/RVC-Boss/GPT-SoVITS/pull/457
|
||||||
|
|
||||||
### 20240214更新
|
## 20240214
|
||||||
|
|
||||||
1-训练支持中文实验名 (原来会报错)
|
1-训练支持中文实验名 (原来会报错)
|
||||||
|
|
||||||
2-DPO训练改为可勾选选项而非必须.如勾选batch size自动减半.修复推理界面新参数不传参的问题.
|
2-DPO训练改为可勾选选项而非必须.如勾选batch size自动减半.修复推理界面新参数不传参的问题.
|
||||||
|
|
||||||
### 20240216更新
|
## 20240216
|
||||||
|
|
||||||
1-支持无参考文本输入
|
1-支持无参考文本输入
|
||||||
|
|
||||||
2-修复中文文本前端bug https://github.com/RVC-Boss/GPT-SoVITS/issues/475
|
2-修复中文文本前端bug https://github.com/RVC-Boss/GPT-SoVITS/issues/475
|
||||||
|
|
||||||
### 20240221更新
|
## 20240221
|
||||||
|
|
||||||
1-数据处理添加语音降噪选项 (降噪为只剩16k采样率, 除非底噪很大先不急着用哦).
|
1-数据处理添加语音降噪选项 (降噪为只剩16k采样率, 除非底噪很大先不急着用哦).
|
||||||
|
|
||||||
@ -135,7 +137,7 @@
|
|||||||
|
|
||||||
4-colab修复不开启公网url
|
4-colab修复不开启公网url
|
||||||
|
|
||||||
### 20240306更新
|
## 20240306
|
||||||
|
|
||||||
1-推理加速50% (RTX3090+pytorch2.2.1+cu11.8+win10+py39 tested) https://github.com/RVC-Boss/GPT-SoVITS/pull/672
|
1-推理加速50% (RTX3090+pytorch2.2.1+cu11.8+win10+py39 tested) https://github.com/RVC-Boss/GPT-SoVITS/pull/672
|
||||||
|
|
||||||
@ -147,7 +149,7 @@
|
|||||||
|
|
||||||
5-修改is_half的判断使在Mac上能正常CPU推理 https://github.com/RVC-Boss/GPT-SoVITS/pull/573
|
5-修改is_half的判断使在Mac上能正常CPU推理 https://github.com/RVC-Boss/GPT-SoVITS/pull/573
|
||||||
|
|
||||||
### 202403/202404/202405更新
|
## 202403/202404/202405
|
||||||
|
|
||||||
2个重点
|
2个重点
|
||||||
|
|
||||||
@ -169,7 +171,7 @@
|
|||||||
|
|
||||||
6-nan自动转fp32阶段的hubert提取bug修复
|
6-nan自动转fp32阶段的hubert提取bug修复
|
||||||
|
|
||||||
### 20240610
|
## 20240610
|
||||||
|
|
||||||
小问题修复:
|
小问题修复:
|
||||||
|
|
||||||
@ -183,7 +185,7 @@
|
|||||||
|
|
||||||
4-修复了webui的GPT中文微调没读到bert导致和推理不一致, 训练太多可能效果还会变差的问题.如果大量数据微调的建议重新微调模型得到质量优化 [#99f09c8](https://github.com/RVC-Boss/GPT-SoVITS/commit/99f09c8bdc155c1f4272b511940717705509582a)
|
4-修复了webui的GPT中文微调没读到bert导致和推理不一致, 训练太多可能效果还会变差的问题.如果大量数据微调的建议重新微调模型得到质量优化 [#99f09c8](https://github.com/RVC-Boss/GPT-SoVITS/commit/99f09c8bdc155c1f4272b511940717705509582a)
|
||||||
|
|
||||||
### 20240706
|
## 20240706
|
||||||
|
|
||||||
小问题修复:
|
小问题修复:
|
||||||
|
|
||||||
@ -203,7 +205,7 @@
|
|||||||
|
|
||||||
后面会逐渐验证快速推理分支的推理改动的一致性
|
后面会逐渐验证快速推理分支的推理改动的一致性
|
||||||
|
|
||||||
### 20240727
|
## 20240727
|
||||||
|
|
||||||
1-清理冗余i18n代码 https://github.com/RVC-Boss/GPT-SoVITS/pull/1298
|
1-清理冗余i18n代码 https://github.com/RVC-Boss/GPT-SoVITS/pull/1298
|
||||||
|
|
||||||
@ -215,109 +217,170 @@
|
|||||||
|
|
||||||
4-[支持合成语速调节.支持冻结随机性只调节语速, ](https://github.com/RVC-Boss/GPT-SoVITS/commit/9588a3c52d9ebdb20b3c5d74f647d12e7c1171c2)并将其更新到api.py上https://github.com/RVC-Boss/GPT-SoVITS/pull/1340
|
4-[支持合成语速调节.支持冻结随机性只调节语速, ](https://github.com/RVC-Boss/GPT-SoVITS/commit/9588a3c52d9ebdb20b3c5d74f647d12e7c1171c2)并将其更新到api.py上https://github.com/RVC-Boss/GPT-SoVITS/pull/1340
|
||||||
|
|
||||||
|
- 2024.07.27 [PR#1306](https://github.com/RVC-Boss/GPT-SoVITS/pull/1306), [PR#1356](https://github.com/RVC-Boss/GPT-SoVITS/pull/1356): 增加 BS-Roformer 人声伴奏分离模型支持.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: KamioRinn
|
||||||
|
- 2024.07.27 [PR#1351](https://github.com/RVC-Boss/GPT-SoVITS/pull/1351): 更好的中文文本前端.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: KamioRinn
|
||||||
|
|
||||||
### 20240806
|
## 202408 (V2 版本)
|
||||||
|
|
||||||
1-增加bs-roformer人声伴奏分离模型支持. https://github.com/RVC-Boss/GPT-SoVITS/pull/1306 https://github.com/RVC-Boss/GPT-SoVITS/pull/1356 [支持fp16推理.](https://github.com/RVC-Boss/GPT-SoVITS/commit/e62e965323a60a76a025bcaa45268c1ddcbcf05c)
|
- 2024.08.01 [PR#1355](https://github.com/RVC-Boss/GPT-SoVITS/pull/1355): 添加自动填充下一步文件路径的功能.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2024.08.01 [Commit#e62e9653](https://github.com/RVC-Boss/GPT-SoVITS/commit/e62e965323a60a76a025bcaa45268c1ddcbcf05c): 支持 BS-Roformer 的 FP16 推理.
|
||||||
|
- 类型: 性能优化
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2024.08.01 [Commit#bce451a2](https://github.com/RVC-Boss/GPT-SoVITS/commit/bce451a2d1641e581e200297d01f219aeaaf7299), [Commit#4c8b7612](https://github.com/RVC-Boss/GPT-SoVITS/commit/4c8b7612206536b8b4435997acb69b25d93acb78): 增加用户友好逻辑, 对用户随意输入的显卡序号也能正常运行.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2024.08.02 [Commit#ff6c193f](https://github.com/RVC-Boss/GPT-SoVITS/commit/ff6c193f6fb99d44eea3648d82ebcee895860a22)~[Commit#de7ee7c7](https://github.com/RVC-Boss/GPT-SoVITS/commit/de7ee7c7c15a2ec137feb0693b4ff3db61fad758): **新增 GPT-SoVITS V2 模型.**
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2024.08.03 [Commit#8a101474](https://github.com/RVC-Boss/GPT-SoVITS/commit/8a101474b5a4f913b4c94fca2e3ca87d0771bae3): 增加粤语 FunASR 支持.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2024.08.03 [PR#1387](https://github.com/RVC-Boss/GPT-SoVITS/pull/1387), [PR#1388](https://github.com/RVC-Boss/GPT-SoVITS/pull/1388): 优化界面, 优化计时逻辑.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2024.08.06 [PR#1404](https://github.com/RVC-Boss/GPT-SoVITS/pull/1404), [PR#987](https://github.com/RVC-Boss/GPT-SoVITS/pull/987), [PR#488](https://github.com/RVC-Boss/GPT-SoVITS/pull/488): 优化多音字逻辑 (V2 版本特供).
|
||||||
|
- 类型: 修复, 新功能
|
||||||
|
- 提交: KamioRinn, RVC-Boss
|
||||||
|
- 2024.08.13 [PR#1422](https://github.com/RVC-Boss/GPT-SoVITS/pull/1422): 修复参考音频混合只能上传一条的错误, 添加数据集检查, 缺失会弹出警告窗口.
|
||||||
|
- 类型: 修复, 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2024.08.20 [Issue#1508](https://github.com/RVC-Boss/GPT-SoVITS/issues/1508): 上游 LangSegment 库支持通过 SSML 标签优化数字、电话、时间日期等.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: juntaosun
|
||||||
|
- 2024.08.20 [PR#1503](https://github.com/RVC-Boss/GPT-SoVITS/pull/1503): 修复并优化 API.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: KamioRinn
|
||||||
|
- 2024.08.20 [PR#1490](https://github.com/RVC-Boss/GPT-SoVITS/pull/1490): 合并 fast_inference 分支.
|
||||||
|
- 类型: 重构
|
||||||
|
- 提交: ChasonJiang
|
||||||
|
- 2024.08.21 **正式发布 GPT-SoVITS V2 版本.**
|
||||||
|
|
||||||
2-更好的中文文本前端. https://github.com/RVC-Boss/GPT-SoVITS/pull/987 https://github.com/RVC-Boss/GPT-SoVITS/pull/1351 https://github.com/RVC-Boss/GPT-SoVITS/pull/1404 优化多音字逻辑 (v2版本特供). https://github.com/RVC-Boss/GPT-SoVITS/pull/488
|
## 202502 (V3 版本)
|
||||||
|
|
||||||
3-自动填充下一步的文件路径 https://github.com/RVC-Boss/GPT-SoVITS/pull/1355
|
- 2025.02.11 [Commit#ed207c4b](https://github.com/RVC-Boss/GPT-SoVITS/commit/ed207c4b879d5296e9be3ae5f7b876729a2c43b8)~[Commit#6e2b4918](https://github.com/RVC-Boss/GPT-SoVITS/commit/6e2b49186c5b961f0de41ea485d398dffa9787b4): **新增 GPT-SoVITS V3 模型, 需要 14G 显存进行微调.**
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2025.02.12 [PR#2032](https://github.com/RVC-Boss/GPT-SoVITS/pull/2032): 更新项目多语言文档.
|
||||||
|
- 类型: 文档
|
||||||
|
- 提交: StaryLan
|
||||||
|
- 2025.02.12 [PR#2033](https://github.com/RVC-Boss/GPT-SoVITS/pull/2033): 更新日语文档.
|
||||||
|
- 类型: 文档
|
||||||
|
- 提交: Fyphen
|
||||||
|
- 2025.02.12 [PR#2010](https://github.com/RVC-Boss/GPT-SoVITS/pull/2010): 优化注意力计算逻辑.
|
||||||
|
- 类型: 性能优化
|
||||||
|
- 提交: wzy3650
|
||||||
|
- 2025.02.12 [PR#2040](https://github.com/RVC-Boss/GPT-SoVITS/pull/2040): 微调添加梯度检查点支持, 需要 12G 显存进行微调.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: Kakaru Hayate
|
||||||
|
- 2025.02.14 [PR#2047](https://github.com/RVC-Boss/GPT-SoVITS/pull/2047), [PR#2062](https://github.com/RVC-Boss/GPT-SoVITS/pull/2062), [PR#2073](https://github.com/RVC-Boss/GPT-SoVITS/pull/2073): 切换新的语言分割工具, 优化多语种混合文本切分策略, 优化文本里的数字和英文处理逻辑.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: KamioRinn
|
||||||
|
- 2025.02.23 [Commit#56509a17](https://github.com/RVC-Boss/GPT-SoVITS/commit/56509a17c918c8d149c48413a672b8ddf437495b)~[Commit#514fb692](https://github.com/RVC-Boss/GPT-SoVITS/commit/514fb692db056a06ed012bc3a5bca2a5b455703e): **GPT-SoVITS V3 模型支持 LoRA 训练, 需要 8G 显存进行微调.**
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2025.02.23 [PR#2078](https://github.com/RVC-Boss/GPT-SoVITS/pull/2078): 人声背景音分离增加 Mel Band Roformer 模型支持.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: Sucial
|
||||||
|
- 2025.02.26 [PR#2112](https://github.com/RVC-Boss/GPT-SoVITS/pull/2112), [PR#2114](https://github.com/RVC-Boss/GPT-SoVITS/pull/2114): 修复中文路径下 Mecab 的报错 (具体表现为日文韩文、文本混合语种切分可能会遇到的报错).
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: KamioRinn
|
||||||
|
- 2025.02.27 [Commit#92961c3f](https://github.com/RVC-Boss/GPT-SoVITS/commit/92961c3f68b96009ff2cd00ce614a11b6c4d026f)~[Commit#](https://github.com/RVC-Boss/GPT-SoVITS/commit/250b1c73cba60db18148b21ec5fbce01fd9d19bc): **支持使用 24KHz 转 48kHz 的音频超分模型**, 缓解 V3 模型生成音频感觉闷的问题.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 关联: [Issue#2085](https://github.com/RVC-Boss/GPT-SoVITS/issues/2085), [Issue#2117](https://github.com/RVC-Boss/GPT-SoVITS/issues/2117)
|
||||||
|
- 2025.02.28 [PR#2123](https://github.com/RVC-Boss/GPT-SoVITS/pull/2123): 更新项目多语言文档
|
||||||
|
- 类型: 文档
|
||||||
|
- 提交: StaryLan
|
||||||
|
- 2025.02.28 [PR#2122](https://github.com/RVC-Boss/GPT-SoVITS/pull/2122): 对于模型无法判断的CJK短字符采用规则判断.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: KamioRinn
|
||||||
|
- 关联: [Issue#2116](https://github.com/RVC-Boss/GPT-SoVITS/issues/2116)
|
||||||
|
|
||||||
4-增加喂饭逻辑, 用户瞎写显卡序号也可以正常运作 [bce451a](https://github.com/RVC-Boss/GPT-SoVITS/commit/bce451a2d1641e581e200297d01f219aeaaf7299) [4c8b761](https://github.com/RVC-Boss/GPT-SoVITS/commit/4c8b7612206536b8b4435997acb69b25d93acb78)
|
- 2025.02.28 [Commit#c38b1690](https://github.com/RVC-Boss/GPT-SoVITS/commit/c38b16901978c1db79491e16905ea3a37a7cf686), [Commit#a32a2b89](https://github.com/RVC-Boss/GPT-SoVITS/commit/a32a2b893436fad56cc82409121c7fa36a1815d5): 增加语速传参以支持调整合成语速.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2025.02.28 **正式发布 GPT-SoVITS V3**.
|
||||||
|
|
||||||
5-增加粤语ASR支持 [8a10147](https://github.com/RVC-Boss/GPT-SoVITS/commit/8a101474b5a4f913b4c94fca2e3ca87d0771bae3)
|
## 202503
|
||||||
|
|
||||||
6-GPT-SoVITS-v2支持
|
- 2025.03.31 [PR#2236](https://github.com/RVC-Boss/GPT-SoVITS/pull/2236): 修复一批由依赖的库版本不对导致的问题.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 关联:
|
||||||
|
- PyOpenJTalk: [Issue#1131](https://github.com/RVC-Boss/GPT-SoVITS/issues/1131), [Issue#2231](https://github.com/RVC-Boss/GPT-SoVITS/issues/2231), [Issue#2233](https://github.com/RVC-Boss/GPT-SoVITS/issues/2233).
|
||||||
|
- ONNX: [Issue#492](https://github.com/RVC-Boss/GPT-SoVITS/issues/492), [Issue#671](https://github.com/RVC-Boss/GPT-SoVITS/issues/671), [Issue#1192](https://github.com/RVC-Boss/GPT-SoVITS/issues/1192), [Issue#1819](https://github.com/RVC-Boss/GPT-SoVITS/issues/1819), [Issue#1841](https://github.com/RVC-Boss/GPT-SoVITS/issues/1841).
|
||||||
|
- Pydantic: [Issue#2230](https://github.com/RVC-Boss/GPT-SoVITS/issues/2230), [Issue#2239](https://github.com/RVC-Boss/GPT-SoVITS/issues/2239).
|
||||||
|
- PyTorch-Lightning: [Issue#2174](https://github.com/RVC-Boss/GPT-SoVITS/issues/2174).
|
||||||
|
- 2025.03.31 [PR#2241](https://github.com/RVC-Boss/GPT-SoVITS/pull/2241): **为 SoVITS v3 适配并行推理**.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: ChasonJiang
|
||||||
|
|
||||||
7-计时逻辑优化 https://github.com/RVC-Boss/GPT-SoVITS/pull/1387
|
- 修复其他若干错误.
|
||||||
|
|
||||||
### 20240821
|
- 整合包修复 onnxruntime GPU 推理的支持
|
||||||
|
- 类型: 修复
|
||||||
|
- 内容:
|
||||||
|
- G2PW 内的 ONNX 模型由 CPU 推理 换为 GPU, 显著降低推理的 CPU 瓶颈;
|
||||||
|
- foxjoy 去混响模型现在可使用 GPU 推理
|
||||||
|
|
||||||
1-fast_inference分支合并进main: https://github.com/RVC-Boss/GPT-SoVITS/pull/1490
|
## 202504 (V4 版本)
|
||||||
|
|
||||||
2-支持通过ssml标签优化数字、电话、时间日期等: https://github.com/RVC-Boss/GPT-SoVITS/issues/1508
|
- 2025.04.01 [Commit#6a60e5ed](https://github.com/RVC-Boss/GPT-SoVITS/commit/6a60e5edb1817af4a61c7a5b196c0d0f1407668f): 解锁 SoVITS v3 并行推理, 修复模型加载异步逻辑.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2025.04.07 [PR#2255](https://github.com/RVC-Boss/GPT-SoVITS/pull/2255): Ruff 格式化代码, 更新 G2PW 链接.
|
||||||
|
- 类型: 风格
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2025.04.15 [PR#2290](https://github.com/RVC-Boss/GPT-SoVITS/pull/2290): 清理文档, 支持 Python 3.11, 更新安装文件.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2025.04.20 [PR#2300](https://github.com/RVC-Boss/GPT-SoVITS/pull/2300): 更新 Colab, 安装文件和模型下载.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2025.04.20 [Commit#e0c452f0](https://github.com/RVC-Boss/GPT-SoVITS/commit/e0c452f0078e8f7eb560b79a54d75573fefa8355)~[Commit#9d481da6](https://github.com/RVC-Boss/GPT-SoVITS/commit/9d481da610aa4b0ef8abf5651fd62800d2b4e8bf): **新增 GPT-SoVITS V4 模型**.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2025.04.21 [Commit#8b394a15](https://github.com/RVC-Boss/GPT-SoVITS/commit/8b394a15bce8e1d85c0b11172442dbe7a6017ca2)~[Commit#bc2fe5ec](https://github.com/RVC-Boss/GPT-SoVITS/commit/bc2fe5ec86536c77bb3794b4be263ac87e4fdae6), [PR#2307](https://github.com/RVC-Boss/GPT-SoVITS/pull/2307): 适配 V4 并行推理.
|
||||||
|
- 类型: 新功能
|
||||||
|
- 提交: RVC-Boss, ChasonJiang
|
||||||
|
- 2025.04.22 [Commit#7405427a](https://github.com/RVC-Boss/GPT-SoVITS/commit/7405427a0ab2a43af63205df401fd6607a408d87)~[Commit#590c83d7](https://github.com/RVC-Boss/GPT-SoVITS/commit/590c83d7667c8d4908f5bdaf2f4c1ba8959d29ff), [PR#2309](https://github.com/RVC-Boss/GPT-SoVITS/pull/2309): 修复模型版本传参.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: RVC-Boss, ChasonJiang
|
||||||
|
- 2025.04.22 [Commit#fbdab94e](https://github.com/RVC-Boss/GPT-SoVITS/commit/fbdab94e17d605d85841af6f94f40a45976dd1d9), [PR#2310](https://github.com/RVC-Boss/GPT-SoVITS/pull/2310): 修复 Numpy 与 Numba 版本不匹配问题, 更新 librosa 版本.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: RVC-Boss, XXXXRT666
|
||||||
|
- 关联: [Issue#2308](https://github.com/RVC-Boss/GPT-SoVITS/issues/2308)
|
||||||
|
- **2024.04.22 正式发布 GPT-SoVITS V4**.
|
||||||
|
- 2025.04.22 [PR#2311](https://github.com/RVC-Boss/GPT-SoVITS/pull/2311): 更新 Gradio 参数.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2025.04.25 [PR#2322](https://github.com/RVC-Boss/GPT-SoVITS/pull/2322): 完善 Colab/Kaggle Notebook 脚本.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
|
||||||
3-api修复优化: https://github.com/RVC-Boss/GPT-SoVITS/pull/1503
|
## 202505
|
||||||
|
|
||||||
4-修复了参考音频混合只能上传一条的bug:https://github.com/RVC-Boss/GPT-SoVITS/pull/1422
|
- 2025.05.26 [PR#2351](https://github.com/RVC-Boss/GPT-SoVITS/pull/2351): 完善 Docker, Windows 自动构建脚本, Pre-Commit 格式化.
|
||||||
|
- 类型: 杂项
|
||||||
|
- 提交: XXXXRT666
|
||||||
|
- 2025.05.26 [PR#2408](https://github.com/RVC-Boss/GPT-SoVITS/pull/2408): 优化混合语种切分识别逻辑.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: KamioRinn
|
||||||
|
- 关联: [Issue#2404](https://github.com/RVC-Boss/GPT-SoVITS/issues/2404)
|
||||||
|
- 2025.05.26 [PR#2377](https://github.com/RVC-Boss/GPT-SoVITS/pull/2377): 通过缓存策略使 SoVITS V3/V4 推理提速 10%.
|
||||||
|
- 类型: 性能优化
|
||||||
|
- 提交: Kakaru Hayate
|
||||||
|
- 2025.05.26 [Commit#4d9d56b1](https://github.com/RVC-Boss/GPT-SoVITS/commit/4d9d56b19638dc434d6eefd9545e4d8639a3e072), [Commit#8c705784](https://github.com/RVC-Boss/GPT-SoVITS/commit/8c705784c50bf438c7b6d0be33a9e5e3cb90e6b2), [Commit#fafe4e7f](https://github.com/RVC-Boss/GPT-SoVITS/commit/fafe4e7f120fba56c5f053c6db30aa675d5951ba): 更新标注界面, 增加友情提示, 即标注完每一面都要点 Submit Text 否则修改无效.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
- 2025.05.29 [Commit#1934fc1e](https://github.com/RVC-Boss/GPT-SoVITS/commit/1934fc1e1b22c4c162bba1bbe7d7ebb132944cdc): 修复 UVR5 和 ONNX 去混响模型使用 FFmpeg 编码 MP3 和 M4A 原路径带空格时的错误.
|
||||||
|
- 类型: 修复
|
||||||
|
- 提交: RVC-Boss
|
||||||
|
|
||||||
5-增加了各种数据集检查,若缺失会弹出warning:https://github.com/RVC-Boss/GPT-SoVITS/pull/1422
|
**预告:端午后基于V2版本进行重大优化更新!**
|
||||||
|
|
||||||
### 20250211
|
|
||||||
|
|
||||||
增加gpt-sovits-v3模型, 需要14G显存可以微调
|
|
||||||
|
|
||||||
### 20250212
|
|
||||||
|
|
||||||
sovits-v3微调支持开启梯度检查点, 需要12G显存可以微调https://github.com/RVC-Boss/GPT-SoVITS/pull/2040
|
|
||||||
|
|
||||||
### 20250214
|
|
||||||
|
|
||||||
优化多语种混合文本切分策略a https://github.com/RVC-Boss/GPT-SoVITS/pull/2047
|
|
||||||
|
|
||||||
### 20250217
|
|
||||||
|
|
||||||
优化文本里的数字和英文处理逻辑https://github.com/RVC-Boss/GPT-SoVITS/pull/2062
|
|
||||||
|
|
||||||
### 20250218
|
|
||||||
|
|
||||||
优化多语种混合文本切分策略b https://github.com/RVC-Boss/GPT-SoVITS/pull/2073
|
|
||||||
|
|
||||||
### 20250223
|
|
||||||
|
|
||||||
1-sovits-v3微调支持lora训练, 需要8G显存可以微调, 效果比全参微调更好
|
|
||||||
|
|
||||||
2-人声背景音分离增加mel band roformer模型支持https://github.com/RVC-Boss/GPT-SoVITS/pull/2078
|
|
||||||
|
|
||||||
### 20250226
|
|
||||||
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/pull/2112 https://github.com/RVC-Boss/GPT-SoVITS/pull/2114
|
|
||||||
|
|
||||||
修复中文路径下mecab的报错 (具体表现为日文韩文、文本混合语种切分可能会遇到的报错)
|
|
||||||
|
|
||||||
### 20250227
|
|
||||||
|
|
||||||
针对v3生成24k音频感觉闷的问题https://github.com/RVC-Boss/GPT-SoVITS/issues/2085 https://github.com/RVC-Boss/GPT-SoVITS/issues/2117 ,支持使用24k to 48k的音频超分模型缓解.
|
|
||||||
|
|
||||||
|
|
||||||
### 20250228
|
|
||||||
|
|
||||||
修复短文本语种选择出错 https://github.com/RVC-Boss/GPT-SoVITS/pull/2122
|
|
||||||
|
|
||||||
修复v3sovits未传参以支持调节语速
|
|
||||||
|
|
||||||
### 202503
|
|
||||||
|
|
||||||
修复一批由依赖的库版本不对导致的问题https://github.com/RVC-Boss/GPT-SoVITS/commit/6c468583c5566e5fbb4fb805e4cc89c403e997b8
|
|
||||||
|
|
||||||
修复模型加载异步逻辑https://github.com/RVC-Boss/GPT-SoVITS/commit/03b662a769946b7a6a8569a354860e8eeeb743aa
|
|
||||||
|
|
||||||
修复其他若干bug
|
|
||||||
|
|
||||||
重点更新:
|
|
||||||
|
|
||||||
1-v3支持并行推理 https://github.com/RVC-Boss/GPT-SoVITS/commit/03b662a769946b7a6a8569a354860e8eeeb743aa
|
|
||||||
|
|
||||||
2-整合包修复onnxruntime GPU推理的支持, 影响: (1) g2pw有个onnx模型原先是CPU推理现在用GPU, 显著降低推理的CPU瓶颈 (2) foxjoy去混响模型现在可使用GPU推理
|
|
||||||
|
|
||||||
### 202504/202505更新
|
|
||||||
|
|
||||||
1-修复uvr5和onnx去混响模型ffmpeg编码mp3和m4a原路径带空格会有bug的问题
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/commit/1934fc1e1b22c4c162bba1bbe7d7ebb132944cdc
|
|
||||||
|
|
||||||
2-标注界面增加友情提示标注完每一面都要点submit text否则白忙活
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/commit/fafe4e7f120fba56c5f053c6db30aa675d5951ba
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/commit/8c705784c50bf438c7b6d0be33a9e5e3cb90e6b2
|
|
||||||
|
|
||||||
3-通过缓存策略使sovits推理提速10%
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/pull/2377
|
|
||||||
|
|
||||||
4-混合语种切分识别逻辑优化
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/pull/2408
|
|
||||||
|
|
||||||
5-完善colab/kaggle notebook脚本,完善linux环境配置脚本,docker环境,windows自动构建脚本
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/commit/ad7df5298bea51273c86c05b5b13f28ed7d9fe16
|
|
||||||
https://github.com/RVC-Boss/GPT-SoVITS/commit/d5e479dad6342222eb4887df627e69c048d2338c
|
|
||||||
|
|
||||||
预告:端午后基于V2版本进行重大优化更新!
|
|
||||||
|
Loading…
x
Reference in New Issue
Block a user