ChatGPT file processing automation (Java version)
SRT file translation may be complex because of possible bad sentence cut, making each text part loosing its context.
ChatGPT, with its ability to handle complex processing, by translating the SRT at once as a whole, can produce much better SRT translations than other available translation tools.
Use the ChatMate Windows Exe release,
or
From de distrib folder, use the launcher for your machine (Java or OpenJDK must be installed).
1- Define a Config name, like SRTtoZH
2- Define a Suffix to add to processed file names, like -ZH
3- Define a Model to use, like gpt-3.5-turbo
4- Define a Part size (~ number of paragraphs to be processed at each ChatGPT call). On each ChatGPT call, the number of input+output tokens is limited. Too many tokens also brings with lower quality result. A value of 30 is certainly a good choice for SRT files. Less than 1 will send the whole file content without cut, with a risk of ChatGPT model context length overload.
5- Enter a valid ChatGPT Key
6- Define a System prompt, like:
Translate all text in Chinese keeping the SRT subtitle format with the sentence cut at best for each numbered section of the original.
1- Copy/Paste a SRT content as a User prompt
2- Click on the Test button
3- After the time needed to process the content, the result should appear on the right
1- Drag and Drop a set of files on the File list on the bottom left
2- Click on the Process all files button
3- Each processed file appears on the right with the suffixed name
https://github.com/EtienneAb3d/karaok-AI
https://github.com/EtienneAb3d/WhisperHallu
https://github.com/EtienneAb3d/WhisperTimeSync
https://github.com/EtienneAb3d/NeuroSpell
https://github.com/EtienneAb3d/OpenNeuroSpell
This tool is a demonstration of our know-how.
If you are interested in a commercial/industrial AI linguistic project, contact us:
https://cubaix.com