Pitch tracking and voiced/unvoiced detection in noisy environment using optimal sequence estimation

Moshe Wasserblat, Mikel Gainza, David Dorran, Yuval Domb

Abstract: This paper addresses the problem of pitch tracking and voiced/unvoiced detection in noisy speech environments. An algorithm is presented which uses a number of variable thresholds to track pitch contour with minimal error. This is achieved by modeling the pitch tracking problem in such a way that allows the use of optimal estimation methods, such MLSE. The performance of the algorithm is evaluated using the Keele pitch detection database with realistic background noise. Results show best performance in comparison to other state of the art pitch detector and successful pitch tracking is possible in low signal to noise conditions.