2012 Volume 2012 Pages 314-319
An algorithm for music transcription system that displays notes and rests on the five lines of music sound is proposed. It is able to process multi-voice music which includes chords, superimposed melodies etc. The system first projects the sound onto the time-frequency plane applying Gabor wavelet transform, and then identifies tempo, the length of notes etc. with referring autocorrelation of the projection. At next, it estimates actually played pitch names that correspond to the fundamental frequencies of simultaneously played tones. It applies a state estimation technique to cope with considerable harmonics which depend upon instruments.
