A method for echo cancellation in audio signals using the genetic algorithm

Abstract

A new method for echo cancellation is proposed. This method is based on inverse models of systems causing the echo in transmission channels. Parameters of these models are optimized using the genetic algorithm. In general, the inverse models that are used for echo cancellation can be divided into several categories according to their complexity. In the simplest case, the models consist of digital delay lines, scalers, and adders. The more complex models use also a nonlinear filtration, which in the general case may be a nonstationary one. The experiments and informal subjective listening tests showed that the proposed method allows for the perceptually efficient suppression of echo which was caused by multiple reflections providing delayed and scaled copies of the original signal. An additional preprocessing employing a calculation of the correlation function of the input signal can diminish the computational complexity of the proposed algorithm. Some examples of echo cancellation results obtained with the use of the proposed method are presented and illustrated with sound examples.