An Approach for Pitch Estimation from Noisy Speech
Celia Shahnaz, Wei‐Ping Zhu, M. Omair Ahmad · 2007
In this paper, a new technique is proposed for the estimation of pitch from the noise-corrupted speech. To enhance speech in a noisy environment, a spectral subtraction (SS) based noise reduction scheme is incorporated prior to pitch estimation. The de-noised speech thus obtained is passed through an inverse filter, whose parameters are derived from the linear prediction (LP) analysis, yielding an output referred to as the LP residual. The direct use of the LP residual which is capable of delivering the knowledge of glottal closure (GC) events, is found to be ineffective for noisy speech. Hence, a new average magnitude sum function (AMSF) of the LP residual is proposed which reveals prominent peaks at the integer multiple of the pitch period even in the presence of noise. Simulation results show that the proposed approach significantly reduces the percentage gross pitch-errors in comparison to that achieved by some of the existing methods both in the white and car environmental noises.