抄録
We propose a cepstrum-based method for pitch estimation of speech signals in noisy environments. The cepstrum method is useful for pitch estimation. The performance of the method, however, degrades in noisy environments. Instead of the speech signal, we employ its windowless autocorrelation function for obtaining the cepstrum. The windowless autocorrelation function is a noise-reduced version of the speech signal where the periodicity is more apparent. Experimental results on male and female voices in the presence of white and color noise show that the gross pitch error becomes lower when using the proposed method compared with other related methods.
