A robust pitch tracking method in noisy environment
Haishan Han, Shasha Xing, Kai Liu, Junli Chen · 2013
A pitch tracking algorithm combining both spectral and temporal method is presented in this paper. The algorithm is robust for speech in various kinds of noisy environment at different signal to noise ratios. In frequency domain the low frequency region energy ratio was computed for voiced and unvoiced determination, correlation of multiple harmonic peaks was computed to find pitch candidate. Low frequency energy ratio and candidates obtained from frequency domain were used to guide the pitch candidate estimation in time domain which was performed both on the filtered speech and the filtered squared speech using normalized cross correlation function. Merit values associate with every candidate computed according to different conditions were computed. Then dynamic programming will used on these pitch merit pairs to find best pitch tracking. Performance of the method on different noisy speech was also evaluated in this paper.