I am running whisper-asr-webservice on my always on pc among other services and today I fed it 14 seconds of noise (I was just batch feeding all recordings from interviews) and this is what whisper generated:

1
00:00:00,000 --> 00:00:03,000
Zdjęcia i napisy stworzone przez społeczność Amara.org

:D

Seems like a lot of polish subtitles that were used to train this model were created with amara