A Frequency-selective Noise Spectral Subtraction Approach

In this paper, a new approach for frequency-selective noise spectral subtraction is proposed, based on the Magnitude Spectral Subtraction (MSS) technique and the two-channel Generalized Sidelobe Canceller (GSC) for noise spectrum estimation. A comparison of the proposed and the conventional noise spectral subtraction techniques is given, based on computer simulations, in terms of noise attenuation performance and the ability to avoid speech distortion. Simulation results clarify that less amount of speech distortion is obtained whereas higher noise attenuation can be achieved, as compared to the conventional noise spectral subtraction technique