Incorrect Transcription Language Output When Specifying Japanese Audio Input

[donwang-ux]

Hello,

I’m working with whisper.cpp and ran into a problem where my Japanese audio file is being transcribed into English, despite specifying the input language as Japanese using the -l ja flag. The command I used is as follows:

main -mc 0 -f input.mp3 -l ja —output-srt -m large-v2.bin

I expected the output to be a transcription in Japanese, aligning with the input audio language setting. However, the transcription returned is in English. I’m wondering if this is an issue with how the language flag is being processed or if I might be missing an additional configuration for handling Japanese audio files correctly.

Could you provide guidance on ensuring the transcription matches the specified input language, or is this potentially a bug that needs to be addressed in the codebase?

Thank you for your help.

[hopto-dot]

The issue you are having is probably to do with Japanese text being in the file name (which is a bug).

Remove all Japanese characters from the file name and it should work.

If not, maybe try using large-v3, that works for me.