A harmonic enhancement based multipitch estimation algorithm

This paper describes a novel computationally efficient method for estimating multiple pitches (multi-F0) in complex audio signals. It is based on an iterative approach, where the fundamental frequency of the predominant pitch is estimated, the detected sound's harmonic structure is then weakened, and the process is repeated for the residual signal. In the estimation stage, an algorithm based on harmonic enhancement is proposed. With those techniques, the method can not only estimate predominant pitch for human sounds and instrumental sounds the same well regardless of their different harmonic structures, but also can robustly estimate multiple pitches. The performance is compared with three recent methods.