Excitation Epoch and Voicing Detection Using Hilbert Envelope with Single-Pass Processing

Abstract

A technique is presented for excitation epoch and voicing detection in speech signal using its Hilbert envelope and employing single-pass processing. The excitation epoch detection comprises dynamic range compression for reducing amplitude variability, Hilbert envelope calculation and dynamic peak detection for excitation saliency enhancement, and epoch marking by locating the maximum-sum subarray peaks. The voicing detection is based on thresholding the inter-epoch similarity calculated as the normalized covariance of the adjacent inter-epoch intervals of the Hilbert envelope. The total algorithmic delay is less than 60 ms. The epoch detection and the voicing detection for clean and telephone-quality speech showed a good match with those obtained from the EGG, and the detection performances compared favorably with the earlier techniques.