Comparison of Time Domain Approaches for Single Channel Speech Enhancement

Abstract

Noise reduction in speech applications like hearing aids, automatic speech recognition, and speech coding is a time-honored problem for decades. Speech is debased by additive background noises, which is to be enhanced by single-channel speech enhancement techniques. The environmental noise created in busy streets, traffic, train, or the cockpit of an airplane becomes a hindrance for speech processing applications. In this paper noise reduction using time-domain approaches like adaptive filtering methods like least mean square(LMS), Recursive least squares (RLS), and Kalman filtering approach are discussed. The quality, Segmental SNR and intelligibility of speech are compared for these approaches. Of those Kalman filter outperforms in aspects of PSEQ, STOI and SSNR.