There are many links in Menu--Help/About, such as model download addresses, CUDA configuration, etc. You can try clicking to use them when you encounter problems.

1. Cannot open after double-clicking sp.exe

The software is developed based on pyside6. The main interface uses a lot of Qt components, and the loading may be slow, ranging from 5 seconds to 2 minutes. Please be patient.

If it still does not display completely within a few minutes, and does not display the startup screen, but only a black window, then the program may have an error. Please check whether there are any errors in the console. For the pre-packaged version, please check whether you have only downloaded the upgrade package. If so, please download the complete package.

If you have tried all methods and waited for a long time but still cannot open it, please open the latest log file in the logs folder, check the error information, or submit the file to GitHub Issue or to bbs.pyvideotrans.com to ask questions and upload the file.

2. Video clarity is reduced

Some operations in the translation process will involve transcoding, and transcoding will inevitably involve quality loss. If you want to minimize the loss, you can perform the following operations:

1. The original video uses libx264 encoded mp4 video
2. In Menu--Tools--Advanced Settings, make the settings as shown in the figure

3. Error during the translation stage

If a red error occurs during the translation process after the subtitles are recognized, this is generally a "network connection error" or an "account problem" with the translation channel used.

If you are using translation channels such as "Google", "Microsoft", "Gemini", etc., then it is likely a network connection problem. You need to enable scientific internet access and fill in the network proxy address provided in the "scientific internet access tool" in the network proxy address text box.

If you are already using scientific internet access, but still report a network connection error, your proxy may be unavailable. Please correctly fill in the http proxy ip and port number provided in the scientific tool.

If you are sure that the proxy is available and still report an error, it is likely that the account is unavailable. For example, Gemini is not available in all countries. You can try switching the proxy node to another country.

Gemini, ChatGPT, and AzureGPT channels generally have request frequency limits. Excluding network problems and account problems, it may be that the request frequency exceeds the limit. At this time, you can open Menu--Tools--Advanced Settings and set the "Pause after translation/s" to 30 or a larger number.

3. Speech recognition accuracy is too low

1. In faster mode and openai mode, using a larger model can improve accuracy. tiny is a small model with poor recognition effect, and large-v3 is the largest model with the best recognition effect. All model download addresses https://pyvideotrans.com/model
2. If the original video is in Chinese pronunciation, you can try using zh_recogn, which has a better effect. Instructions for use https://pyvideotrans.com/zh_recogn.html
3. Select "Keep background sound" to denoise in advance, and the recognition effect will be better, but note that if the video is very large, do not select it.

4. Model download address

Model download address https://pyvideotrans.com/model

5. Is it available on Win7?

Win7 system is not supported

6. Prompts that python310.dll is missing

You may have only downloaded the upgrade patch package. The patch package cannot be used alone. Please download the complete 1.9G package first, and then download the patch package to overwrite it after decompression.

7. Error during the merging stage ffprobe {}

It is likely that the spaces or single and double quotes in the original video name caused the error. Try renaming the original video. For example, the original name is D:/UNSW/2024 T2/BIOS 2061/Week 5 Amphibians, reptiles, and birds/_video_out/BIOS2061-5246_00069- Lecture 13 - Birds 1 'Origin of Birds' - Prof. Richard Kingsford- Part 2 - UNSW##BIOS2061-5246_00069- Lecture 13 - Birds 1 'Origin of Birds' - Prof. Richard Kingsford- Part 2 - UNSW.mp4. You can see that the video name is very complicated, with spaces, single quotes, etc. This kind of name is very easy to make mistakes during processing. Please delete spaces and single quotes.

Does it support docker deployment?

Not supported

Can it recognize the subtitle text in the video, that is, OCR recognition?

The principle of this software is to recognize human speech sounds in the video and convert them into text subtitles. It does not support OCR recognition subtitle function.

Can it be called through the http api interface?

Not currently, but this function may be added later.

Can new languages be added?

No, because speech recognition for subtitles depends on the whisper model, and the languages supported by this model are limited. Unsupported languages will not be recognized.

Where to download the software

https://pyvideotrans.com/downpackage.html

Where to download the model

https://pyvideotrans.com/model.html

CUDA is already installed, but it still cannot be used

Possible reasons:

1: Built-in CUDA support requires version 11.8 or above. Check whether your CUDA version is too low. 2: The graphics card driver is too old and needs to be updated. 3: cudnn is not installed. 4: The graphics card is not an N card or is incompatible.

Does it support multi-role recognition and dubbing?

Not supported. The recognized subtitles do not distinguish between speakers and roles. This function can be manually implemented through "Set Line Role".

The cli command line mode always has problems

The cli mode update is delayed, please use the old version.

Error during the translation stage

Please change the translation channel or fill in the network proxy.

The software freezes after double-clicking and cannot start, stuck on the startup screen

The software is large, please wait patiently. If it still cannot be opened for a long time, please try:

1. Close anti-virus software, security software, etc.
2. Confirm that the path and directory where the software is located are composed of English or numbers, and do not contain spaces, Chinese, special symbols, etc.

If it cannot be started after overwriting the upgrade package, please download the complete package.

If it is already a complete package, please wait patiently. If it still does not start after 2 minutes, please try to force close and reopen.

What translations are supported?

Currently supports Microsoft Translator Google Translate Baidu Translate Tencent Translate DeepL Translate ChatGPT Translate AzureGPT Translate Gemini Pro Translate DeepLx Translate OTT Offline Translate FreeGoogle Translate FreeChatGPT Translate

Connection error

If a Connection error occurs, it means that the network connection failed. If you have not filled in the proxy in the software interface, please fill it in. You cannot directly connect to ChatGPT/Gemini/Google APIs in China. You need to fill in the proxy. Note that being able to open the corresponding website in the browser does not mean that it can be used in the software. Please fill in the correct proxy address in the network proxy input box

Whole all out of memory

If a Whole all out of memory error occurs, it means that the video memory is insufficient. Please use a smaller model, such as tiny small, etc.

Requested float16 compute type, but the target device or backend do not support efficient float 16 computation

The error shows that the current graphics card does not support this data type. The solution is to open Menu--Tools--Advanced Settings and find

CUDA data type

This line, change the content to int8_float16

Then restart the software and execute it. If the error still occurs, modify it to

float32

How to install

No installation is required. After downloading the complete package and decompressing it, double-click sp.exe to use it.

Why is it reported as a virus or blocked?

This is software packaged using pyinstaller. It has not been digitally signed or certified by anti-virus software, and may be misreported. Please add it to the trusted whitelist or close the security software. Or use source code deployment

What TTS voices are supported?

edgeTTS / Azure AI / GPT-SoVITS / clone-voice / elevenlabs

Source code deployment problem description

1. The default is to use ctranslate2 version 4.x, which only supports CUDA12.x. If your CUDA is lower than 12 and you cannot upgrade CUDA to 12.x, please execute the command to uninstall ctranslate2 and then reinstall it.

pip uninstall -y ctranslate2

pip install ctranslate2==3.24.0

2. You may encounter errors such as xx module not found. Please open requirements.txt, search for the xx module, and then remove the == after xx and the version number after it.

Is there any customer service?

No, this is free software, there is no income and no profit, so it cannot be equipped with customer service.

Is there a fee?

This is a free and open source project, and no fees are charged for free use. The translation and tts interfaces are charged by the respective api merchants, which has nothing to do with this project.

Can it be used commercially?

Individuals or companies can use it casually. However, if you want to integrate it into commercial projects, please follow the GPL-v3 open source agreement.