What the audio alone says was played: one event per detected attack with
the notes heard and their confidence.
The spectrum is read a few times just after each attack (skipping the
broadband click of the attack itself) and averaged before pitches are
estimated. Attacks that are not tones — a knock, a click, a chair — have a
flat spectrum and are skipped.
What the audio alone says was played: one event per detected attack with the notes heard and their confidence.
The spectrum is read a few times just after each attack (skipping the broadband click of the attack itself) and averaged before pitches are estimated. Attacks that are not tones — a knock, a click, a chair — have a flat spectrum and are skipped.