Any luck fine tuning for multiple languages?

#113
by gcasey2 - opened

Has anyone here had any success fine tuning whisper on multiple languages. I'm trying to train the model to work on audios that contain two different languages. By default, whisper usually ends up translating the majority of one language into the other one. I've tried fine tuning using different methods(translation with bleu, transcription with mixed dataset) and I cannot seem to get any good results. I feel like this problem should be more common. Appreciate any advice

Sign up or log in to comment