An innovative approach to transcription of polyphonic signals.

Abstract

In this paper, a new method is proposed to separate and identify polyphonic signals generated by an acoustic piano. The method consists of a top down extraction architecture using a harmonics-tracking algorithm followed by a bottom up reconstruction architecture based on a set of the expected profiles of single notes. In addition to the architectures mentioned above, a heuristic function is also included to further re-enforce the robustness and the ability to recognize multiple notes and to pull apart octave unison. The algorithm was tested with chords that consist of multiple notes. Some of the notes are one octave apart. It is shown that a recognition accuracy of 96% can be achieved using the proposed method.