How did you get around the native Mac transcription being terrible and whisper being slow? Did you take the approach most do where you approximate with dictation first and then hot swap text a few seconds later after it’s processed through a better model?