4 ms·

I think he means that even whisper segments the audio into 30 second bits and does transcribing on them and then stiches everything together.