OpenAI’s Whisper can transcribe spoken audio from English and 96 other languages, and can then translate to English text. Not yet a public service, but it can’t be long now — as there’s a new technical paper and open source-code under an MIT licence.
It appears that these are the languages it performs best with…
I found specialists independently saying, elsewhere, that 8% is a “quite decent” score for auto-translation from clearly enunciated and well-recorded audio. So above you have the range roughly either side of that 8% figure. Of course the average lecture-hall recording or iffy phone+skype podcast interview may do significantly worse.
But generally the makers hail the “high accuracy and ease of use”, compared to other methods.