A novel end‐point detection method for noisy speech signal

This paper presents a novel and robust end‐point detection algorithm in the presence of noise. Zero crossing rate and energy of the speech signal have been the most widely used features for locating the end point of an utterance. Most detection algorithms with these parameters work quite well in the high signal‐to‐noise ratio environment, but they show very poor performance with background noise, i.e., in the low signal‐to‐noise ratio environment. Therefore robust end‐point detection under a noise condition still remains an unsolved problem. A new detection measure based on a discrete wavelet transformed speech signal is proposed. As a detection parameter, the sum of standard deviations of wavelet coefficients in the first and weighted fourth scale are used. Since the new feature parameter enhances high‐frequency components of the weak fricative sounds, better discriminating between speech segments and background noise is possible. Experimental results and examples demonstrate that the proposed method is ...