A selective algorithm for the reduction of irregular noise in speech communication

The properties of noise signals can be the main problem associated with noise cancellation systems. In order to overcome this problem, high complexity algorithms have to be used in order to reduce the noise embedded in the useful signals such as speech. This method can be impractical, especially in real-time applications where the computational power is a crucial issue. Adaptive filters give applicable solutions, but most literature proposed a single, yet complex algorithm to removing the noise. This paper proposes an alternative approach to eliminate background noise in corrupted speech signals. The method is achieved by letting the system assigns an appropriate algorithm according to the characteristics of the noise. The criterion used here is based on the calculation of eigenvalue spread in the autocorrelation matrix of the input signal. In addition, an algorithm derived from set-membership filtering is also used among the selected algorithms. This approach showed its potential capability in eliminating different types of environmental noise from corrupted speech signals. The technique presented here exhibited fast convergence speed and improvement in signal-to-noise ratio compared with other single adaptive algorithms.