Pitch Detection Algorithms and Voiced/Unvoiced Classification for Noisy Speech
Ekaterina Verteletskaya, Kirill Sakhnov, Boris Šimák · 2009
This paper describes pitch tracking techniques, which combine voiced/unvoiced classification and pitch estimation based on cepstral analysis, time autocorrelation, spectro-temporal autocorrelation (STA) and average magnitude difference function (AMDF). Pre- and post processing techniques improving performance of pitch detection algorithms (PDAs) are also presented. PDAs have been evaluated by telephone speech signals, corrupted by additive noise, in order to provide comparison and demonstrate their performance and robustness. Speech signals used for evaluation were taken from the Czech telephone speech database consisted of 5 male and 5 female speakers.