A Pitch Estimation Method Robust to High Levels of Noise

Abstract

Pitch is one of the most key parameter in speech coding, speech synthesis and so on, the traditional methods for pitch detection are prone to error at a low SNR at present. A pitch detection method based on pitch harmonic (PH) and the harmonic number based on PH is proposed in this paper. At first, the pitch harmonic is roughly estimated by pitch estimation filter with amplitude compression (PEFAC). Secondly, the weighted algorithm based on modified circular average magnitude difference function (MCAMDF) and pulse sequence is used to compute the pitch harmonic number. At last a pitch tracking method is applied to compute the pitch period candidates accurately. By simulation experiments, it is shown that the proposed pitch detection method has more accurate and more low algorithm complexity than the traditional methods at both high and low SNR.