can basic-pitch provide lower level api than predict() with audio buffer as input ? real time transcription looks cool.