On extracting pitch from noisy speech signals based on spectral and temporal enhancement

A new pitch extractor based on spectral and temporal enhancement of noisy speech signals is presented in this paper. A discrete cosine transform based modified power spectral subtraction scheme is developed and employed prior to pitch extraction in order to suppress the underlying non-stationary noise. The de-noised speech thus obtained is then passed through an inverse filter, whose parameters are derived from the linear prediction (LP) analysis, yielding an output referred to as the LP residual. Since the LP residual is capable of delivering the knowledge of Glottal Closure instants, exploiting its high correlation property, an average magnitude sum function as well as an average magnitude difference function are introduced. The periodicity property of both the functions is argued to be integrated and an enhanced temporal function is put forward for robust pitch extraction in a multi-talker babble noise scenario. The superior efficacy of the proposed pitch extractor relative to some of the existing ones is confirmed through simulation results using the Keele database.