Low distortion SNR-based speech enhancement employing critical band filter banks

This paper presents a speech enhancement method for personal communications where the input signal is divided into a number of subbands that are individually and adaptively weighted in time domain according to a short term SNR estimate in each subband at every time instant. Hence the name adaptive gain equalizer. The method is focused on speech enhancement, acting as a speech booster, and remains idle when the SNR in a particular subband is low. Hence, background artifacts are eliminated and there is no need for a voice activity detector (VAD). This paper also presents a method for least squares design of nonuniform filter banks for application in subband signal processing. Design objectives aim to optimize the filter bank frequency response while minimizing subband and output aliasing. Critical band filter banks with increasing bandwidth are designed and evaluated with the speech enhancement method.