A Pitch Detection Algorithm Based on Windowless Autocorrelation Function and Modified Cepstrum Method in Noisy Environments

This paper proposes a new pitch detection algorithm of speech signals in noisy environment. The performance of the cepstrum method is effected due to the formant effect and the presence of spurious peaks introduced in noisy condition. In our proposed method, we firstly employ windowless autocorrelation function instead of its speech signal for obtaining the cepstrum. The windowless autocorrelation function is a noise-reduced version of the speech signal where the periodicity is more apparent with enhanced pitch peak. Secondly the modified cepstrum method is applied to windowless autocorrelation function which utilizes clipping and band pass filtering operation on log spectrum. The performance of the proposed pitch detection method is compared in terms of gross pitch error with the other related methods. Experimental results on male and female voices in white and color noises shows the superiority of the proposed method over some of the related methods under low levels of signal to noise ratio.