v3.19

jianchang512/pyvideotransv3.19Nov 23, 2024by jianchang512

AI Summary

Adds Bengali language support, fixes volume adjustment logic across all channels, and improves voice cloning normalization. Also adds VAD segmentation for Gemini recognition.

Key Highlights

  • Bengali language support
  • Volume adjustment fix for all channels
  • Voice cloning normalization
  • VAD segmentation for Gemini recognition

New Features

  • Bengali language support
  • VAD segmentation for Gemini

Full Release Notes

> 预打包版仅适用于Windows10/11,MacOS和Linux请源码部署

- Fix: 音量调整对所有渠道均生效,数值代表增减百分比
- Fix: stt语音识别渠道提示不支持的格式错误
- Fix:  #639   #630 #636
- Fix: 克隆声音时对中文英文 文本规范化
- Fix: 试听配音报错提示
- Feat: 添加孟加拉语
- Fix: 单个视频翻译中途停止时清空字幕区
- Fix: 默认选中语音降噪
- Feat: Gemini语音识别时增加vad切割为小片段再挨个发送请求识别,以避免时间戳不准确问题,翻译设置Gemini pro中启用(免费Gemini账号勿启用,会导致大量429错误)


## Win v3.24 完整包下载3.5G(含tiny/medium模型)

> 如果未安装过旧版本,请在此下载完整版,如需cuda加速,需有英伟达显卡并且安装cuda12.x及cudnn9

百度网盘下载地址:    https://pan.baidu.com/s/19c2HrofGQy3KSJayLWNBtw?pwd=cx5v

Huggingface(墙外地址):  https://huggingface.co/spaces/mortimerme/s4/resolve/main/win-videotrans-v3.24.7z?download=true

 

## v3.25补丁包165MB

> 如果已安装过2.x版本,可下载补丁包后解压在sp.exe所在目录,覆盖已有sp.exe和文件夹

GitHub地址: https://github.com/jianchang512/pyvideotrans/releases/download/v3.19/win-PatchUpdate-3.25.7z

百度网盘下载地址:   https://pan.baidu.com/s/1151v1KVXb1wnadTKHykAVA?pwd=bpvt


> 若补丁覆盖后打开出错,请升级显卡驱动、升级cuda到12.x、升级cudnn到cudnn9,并重新下载完整包,然后再使用该补丁包覆盖。
> 
> 或下载该压缩包 https://pan.baidu.com/s/1LOjr9ampzqp7eBrjNgSTHQ?pwd=wrkn   解压后(也可从此地址下载  
 https://github.com/jianchang512/pyvideotrans/releases/download/v3.15/cuda-torch-pathupdate.7z  ),复制其内的 `_internal`文件夹覆盖软件目录内的`_internal`文件夹 

----

> api.exe 下载,api.exe是用于通过http请求调用的文件,不可独立使用,需先下载完整包,然后下载api.exe和 sp.exe放在同一目录下
> 
> api.exe下载地址:https://github.com/jianchang512/pyvideotrans/releases/download/v3.00/api.exe


----

### 所有模型下载地址

> 为避免压缩包体积过大,预打包版只内置最小模型 tiny,识别效果不佳,效果更好的模型请点击下载  

https://github.com/jianchang512/stt/releases/tag/0.0