Pitch detection and voicing/unvoicing decision of Arabic speech signal by HOS-polycesptre

One of the most important issues in Speech Analysis is the extraction of a set of parameters that can, hopefully, help understand the speech process. Depending on the nature of the sound, these parameters could be the fundamental frequency (Pitch) and the harmonics (or formants). The first and second order statistics (autocorrelation, power spectrum, cepstrum) have proved its efficiency for detecting the pitch in signals without noise. The performances diminish, however, when signals are corrupted with noise. Research targeted towards development of methods that can resist to noise. In recent years new tools have been developed referred to as higher order statistics (HOS). These statistics have solved problems that have not been resolved by the first and second order statistics and to extract more information from the signal. In this article we use higher order statistics, in particular polycepstrale analysis for the detection of the fundamental frequency (pitch).