How to use the language parameter of WhisperParams

[martinlexow]

This is more of a question than an issue: How do I use the language parameter correctly, and what should/does it do?

Unfortunately, I can’t find an explanation for it in the documentation.

I had assumed that by setting a language, I could specify the language in which my audio file was recorded (i.e. to improve the overall transcription quality). However, during my tests, I found that the parameter also translates transcripts into the specified language.

I’d really appreciate any explanation you could share! 🙏

[martinlexow]

After looking at the whisper.cpp repo I think it’s:

-l LANG, --language LANG [en] spoken language ('auto' for auto-detect)

So, if the model translates the audio file, this would be an unwanted side effect?