Recovering Overlapping Partials for Monaural Perfect Harmonic Musical Sound Separation Using Modified Common Amplitude Modulation

摘要

Monaural musical sound separation attempts to isolate one or more instrument sources from a mono-channel polyphonic mixture. The primary challenge is to accurately separate pitched musical sounds where their partials overlap with each other. In the general case, each source has one or more non-overlapping partials. However, there exists the case that no non-overlapping partial is available for at least one source, leading to a more complex problem. In order to address this problem, we propose a novel method named Modified Common Amplitude Modulation (Modified-CAM), which aims to separate monaural perfect harmonic musical sound (except the unison), especially for the perfect octave interval. The proposed system can accurately recover both the amplitudes and the phases of the perfect octave, which can be considered as the typical but the hardest condition. We evaluate the proposed system on the University of Iowa musical instrument sample database, and the experiment results show the system's superiority.