Nature Inspired Algorithms in Adaptive Filters used for Speech Enhancement: A Review

Authors

Parvinder Kaur

Research Scholar, Deptartment of Electronics and Communication Engineering, M.M.University Mullana

Jyoti Goel

Deptartment of Electronics and Communication Engineering, Chandigarh College of Engineering and Technology, Chandigarh

Abstract

Speech enhancement aims to improve the quality and intelligibility of speech communication systems in noisy environment. Speech enhancement is important for the communication devices such as mobile phones and hands free telephones. Speech enhancement techniques are classified as single channel and multi channel techniques [1]. This classification is based on no of microphones used to collect the acoustic signal and noise. Single channel noise estimation techniques depend on the spectral characteristics of noisy speech signals. So, it cannot reliably estimate the non-stationary noise. Moreover, single channel noise estimation suffers from noise either due to under estimation or over-estimation during the active region of speech or both. Dual channel adaptive noise cancellation (ANC) is an alternative way to remove the noise from the signal. In ANC, correlated noise signal is adaptively filtered to minimize the output power between the two channels. Dual channel algorithms works on spatial characteristics of speech and noise that can be used for noise removal. So far different gradient based algorithms have been designed to adjust the adaptive filter weights in dual channel systems. Some of the gradient based algorithms are least mean square (LMS), Normalized least mean square (NLMS), Recursive least square (RLS) [2]- [4]