Finding Pitch Markers using First Order Gaussian Differentiator

Abstract

In this paper we propose a method for detecting pitch markers in the speech signal using first order Gaussian differentiator. The speech signal is processed by the linear prediction (LP) analysis to extract the LP residual signal. The peaks around the glottal closure instants in the LP residual are used as pitch markers in this study. The LP residual is convolved with the first order Gaussian differentiator of length 20 msec. Due to the anti-symmetric nature of the first order Gaussian differentiator, there are zero-crossings around the significant peaks in the LP residual. Some of the detected zero-crossings may also correspond to peaks due to excitations like glottal openings in voiced speech and bursts and friction in unvoiced speech. These unwanted zero-crossings are eliminated using energy of the LP residual. The remaining peaks are hypothesized as pitch markers in the speech signal. The proposed method works for both male and female speakers, but for clean speech case only