This has changed:
- new pitch estimation algorithm that also works for singing, not only whistling (it uses autocorrelation for looking at the periodicity, not FFT for looking at the spectrum, and is thus less easily confused by overtones from singing)
- new note splitting algorithm. Splits notes based on: presence/absence of pitch (if there is a gap in pitch, a new note starts), intensity (if one gets quiet and loud again, a new note starts even if the pitch stays present), and pitch (if the pitch changes dramatically, a new note starts even if the intensity stays high).
- a graph shows the recognized pitches and intensities; if things go wrong, this should help with assessing whether the note splitting algorithm is stupid or whether you simply gave it a shitty recording and thus it never had a chance.