An Adaptive Learning Based Speech Enhancement Technique for Communication Systems

摘要

Extraction of speech signal from contaminated signal is main task in all practical applications. While transmitting speech signal, many undesired components are added to desired speech signal and they are eliminated at destination by using adaptive algorithms. Conventional least mean square (LMS) algorithm is widely used because of its simplicity and robustness, step size is main parameter in this algorithm. If there is rapid increase in step size it will affect convergence rate and mean square error (MSE). There is a tradeoff between MSE and convergence. With variable step size, performance of algorithm is improved. Hence developed data variable, error variable, step variable and time variable based adaptive algorithms are proposed. Based on LMS, several adaptive noise elimination techniques are proposed and they are analyzed. In these algorithms step size is variable instead of constant step size and it is based on error signals at particular instant. By proposed algorithm it improves speech signal so that MSE is reduced further signal to noise ratio is also improved.