A new method of fundamental frequency extraction in frequency domain

Abstract

This paper proposes a novel fundamental frequency (pitch) extraction method in the frequency domain. The procedure of the algorithm is as follows. The harmonic peaks of the short-time spectrum of the speech signal are extracted by a binary inhibition network. Then the pitch is determined by the average interval of the harmonics. The method is evaluated on the COSDIC speech database. For the purpose of comparison, the same results for the cepstrum- and autocorrelation-based pitch extraction methods are also presented. The results show that the new method is superior to the cepstrum-based and autocorrelation-based ones. Moreover, it is more robust to white noise than the other two methods.