Publication

Abstract : Epoch in speech, represent the instant where maximum excitation at the vocal tract is obtained. Existing epoch extraction algorithms are capable of accurately extracting epoch information from clean speech signals. However, epoch extraction of band limited signals such as telephonic speech is challenging due to the attenuation of the fundamental frequency components. The present work is focused on improving the performance of epoch extraction from telephonic speech signals by exploiting the properties of Chebyshev polynomial interpolation and by reinforcing the frequency components around the fundamental frequency through the Hilbert envelope (HE). The proposed algorithm brings a refinement of the existing Zero Frequency Filtering (ZFF) method by incorporating Chebyshev interpolation. The proposed refinements to the ZFF algorithm confirmed to provide improved epoch identification rate, identification accuracy, reduced miss rate and false alarm rate. The epoch identification rate of the proposed method is observed to be better than existing methods like Dynamic Programming Phase Slope Algorithm (DYPSA), Speech Event Detection using the Residual Excitation And a Mean-based Signal (SEDREAMS), Dynamic Plosion Index (DPI) and Single Pole Filtering (SPF) methods for telephonic speech quality. © 2018 International Speech Communication Association. All rights reserved.