Pitch Estimation Based on the Cepstrum Analysis by the Multi Scale Product of Clean and Noisy Speech

Abstract

In this paper we propose a new method for estimating the pitch from the speech signal which consists of analysing real cepstrum by the multiscale product (MP) using continuous wavelet transform (WTC) having one vanishing moment (CAMP). Our approach to estimate the pitch consists of the following steps: first we frame the voiced signal, second we calculate the real cepstrum of each frame. Finally, we compute the MP of the cepstrum. The MP is the product of the WTC at three scales. Our method will be evaluated by the Keele database under clean and noisy conditions. Experimental results indicate that the gross pitch errors (GPE) are lower than the compared methods under clean and noisy conditions.