A weighted autocorrelation method for pitch extraction of noisy speech

Abstract

Pitch period (or fundamental frequency) extraction plays an important role on speech processing and has a wide spread of applications in systems associated with speech. Many pitch extraction methods have been proposed so far, but improvement in noisy environments is still a remaining subject. In this paper, we propose a modified version of the autocorrelation method which is well known to be robust against noise. Utilizing that the difference function (amplitude difference function) has similar characteristics with the autocorrelation function, the autocorrelation function is weighted by the reciprocal of the difference function. By simulation experiments based on continuous speech, it is shown that the proposed pitch extraction method behaves more robustly than the conventional methods against additive noise, and especially it is very effective at low signal-to-noise ratio.