Introduction

Abstract

Speech enhancement is the most efficient and convenient technique for improving the quality of communication between human beings in a natural way. The speech communication devices, such as cellular phones, the speech recognition devices such as hand-free input systems for voice dialling and voice-activated security systems utilize the speech processing systems to communicate and store the speech signals. When these systems are in noisy environment, the additive background noise significantly degrades the performance of these systems resulting in inaccurate information exchange and listener fatigue between the speaker and the listener. Hence, speech enhancement is incorporated in voice communication devices to enhance the degraded speech. Depending on the specific applications, the goal of the speech enhancement varies and it could be to reduce the listener fatigue, to enhance the overall speech quality, to increase the intelligibility and to improve the performance of the voice communication device. The reduction of background noise in speech signal may introduce speech distortion in the enhanced signal which in turn may reduce the intelligibility of the enhanced speech. Hence, for the researchers, the main challenge is to develop the speech enhancement algorithms to improve the quality of the speech signal without reducing its intelligibility. Speech enhancement strategies are categorized as single-channel and multichannel enhancement techniques (Loizou in Speech enhancement theory and practice. CRC press 2007), depending on the number of microphones used to collect the acoustic signal and noise. Lot of work has been done in single-channel enhancement, whereas dual-channel enhancement is not extensively studied. It is well known that single-channel speech enhancement cannot reliably estimate the non-stationary noise such as babble noise in which the spectral characteristics vary rapidly in time. Moreover, single-channel noise estimation suffers from noise underestimation and/or overestimation during speech active region. An alternative approach for noise reduction is the dual-channel adaptive noise cancellation (ANC) in which a correlated noise source is adaptively filtered to minimize the output power between the two microphones. This book focuses on the dual-channel enhancement method.