Modified Spectral Subtraction Method for Enhancement of Noisy Speech

This paper proposes a modified spectral subtraction method for enhancing speech corrupted by additive background noise. The method is based on identifying and enhancing speech regions in the noisy speech signal. The speech regions are identified by computing the kurtosis and energy values from the noisy speech signals. For speech corrupted by additive background noise, kurtosis and energy values will be relatively high in speech regions compared to nonspeech regions. This property is exploited in deriving a weight function. The noisy signal is multiplied by the weight function to attenuate the noise in the nonspeech regions. The weighted speech signal is processed only in the speech regions by the spectral subtraction method to enhance speech components in the detected speech regions. The proposed method is effective in attenuating the noise components compared to the spectral subtraction method especially in nonspeech regions.