Yup, we are using Whisper to transcribe automatically so we can train the model on just speech recordings, without human transcripts.

This works for any language that is well supported by the OpenAI Whisper model.