Musical Fundamental Frequency Estimator Based on Harmonic Pattern Match

Abstract

The fundamental frequency (F0) plays a vital role in music signal analysis and processing. However, due to non-stationary noise, undesired physical vibration from the musical instruments, the robust estimation of F0 remains a main challenge. In this paper, a F0 estimation algorithm of music signals based on harmonic pattern match (HPM) is described to achieve more reliable estimation accuracy. The algorithm utilizes the autocorrelation both in the time domain and in the frequency domain, exploiting the spectrum subset to guide the search of F0 candidates (FCs), and an efficient mechanism to evaluate the match between each FC and the harmonic pattern of the input signal. The harmonic pattern of the measured spectrum is presented by sub-pitch in each segmented sub-band. Finally, the estimated $$ \hat{F}_0$$ is selected to best match the sub-pitches under a weighting strategy. Evaluation experiments were performed over a musical instruments database consisting of single pitched notes and the viability of the HPM algorithm was demonstrated to be competitive with several other F0 estimators.