Speech enhancement by using novel multiband spectral subtraction method along with a reduction of the cross spectral component

Abstract

It is essential to enhance the speech signal's clarity and quality in order to maintain the message's content. By boosting the noisy voice signal, the speech signal quality can be raised. Two techniques are presented in this study to significantly minimize the additive background noise. In order to minimize non-stationary additive noise concerning the speech signal, the first approach employs modified multiband spectral subtraction. With this technique, spectral subtraction is carried out based on the signal to noise ratio (SNR) values in various noisy speech frames. When the noisy signal and noise signal are somewhat correlated, a second method is used to minimize the cross spectral components. These techniques are used to get over the drawbacks of the fundamental spectrum subtraction method. To improve the noisy speech signal, both techniques are combined.