A fundamental frequency extraction method based on windowless and normalized autocorrelation functions

This paper presents a fundamental frequency estimation algorithm of noisy speech signal using the combination of windowless and normalized autocorrelation functions. Instead of the input speech signal, we employ its windowless autocorrelation function for obtaining the normalized autocorrelation function. The windowless autocorrelation function is a noise-reduced version of the input speech signal where the periodicity is more apparent with enhanced pitch peak. Experimental results on male and female voices in white noise indicate that the proposed method sufficiently outperforms existing methods in terms of gross pitch error.