from Hacker News

OpenAi Whisper undocumented translation into other languages?

by roberdam on 10/17/22, 10:55 AM with 0 comments

I am doing some research and using Whisper to transcribe audio in spanish, on several of the recordings when someone speaks portuguese by chance in the first 30 seconds and it recognizes the language as portuguese, Whisper continues to transcribe the audio in perfect portuguese even though the original audio is in spanish.

I also tried when transcribing spanish audio selecting portuguese as the transcript language, and it translated to portuguese with almost no errors.

It may be because of the similarity between spanish and portuguese that works so well?, does anyone else care to check this feature in other languages?