A method for pitch estimation from noisy speech signals based on a pitch-harmonic extraction

Abstract

A new method for pitch estimation from noisy speech signals based on a pitch-harmonic extraction is presented in this paper. At first, a variable-length average magnitude difference function (VLAMDF) that is able to alleviate the adverse effects of intrinsic falling minima of the conventional AMDF, has been proposed. We argue that the fast Fourier transform (FFT) power spectrum of the VLAMDF is capable of revealing an estimate of a pitch-harmonic (PH) more accurately even in a heavy noisy scenario. Then, exploiting the extracted PH, a two-fold criterion is developed in the FFT domain to acquire the corresponding harmonic number for the determination of the pitch. The superior efficacy of the proposed pitch estimator relative to some of the existing ones is confirmed through simulation results using the Keele database.