4 ms·

I think the point is having it for real-time; this is for conversations rather than transcribing audio files.

• jamilton 8mo ago
  That quote was for the non-realtime model.