Yeah I checked that out. My current Whisper model is small.en which has a lower WER and is fast enough for real time transcription.