Audio coding for conversion to MIDI

We concentrate on the problem of converting audio samples into MIDI data, and describe a technique for processing music signals that attempts to extract note onsets and pitches, for a small class of music signals. Segmentation of the signals is based upon common onsets of tones and their proximity to edges in the signal. Pitch detection is performed by decomposing the short-time spectra of the segments as linear combinations of spectra in a dictionary of recorded sound segments.