High Order statistics Based pitch Detection of Speech with Additive Gausian Noise

摘要

A new method is porposed for the pitch detection of the noisy speech frames. It is based on combining the widely used average magnitude difference functio with higher-order statistics, namely sum-of-cumulants. We show by comparison that this method performs better than the conventional ones when the speech signal is highly corrupted by additive gaussian noise. The effective pitch detection for noise corrupted speech data has important implication for speech quality enhancement as well as speech recogintion.