MTF based Kalman filtering with linear prediction for power envelope restoration

Abstract

A method based on modulation transfer function (MTF) has been developed to restore the power envelope of noisy reverberant speech. Its advantage is that it can simultaneously suppress the effect of noise and reverberation by restoring the smeared MTF without measuring room impulse responses. The proposed method has power envelope subtraction and power envelope dereverberation. The power envelope subtraction can only reduce the mean value of the noise power envelope while the fluctuations still remain, which greatly affects the dereverberation process. We proposed MTF based Kalman filtering with linear prediction with/without a training phase to remove the fluctuations of the noise power envelope. Objective experiments were conducted in various noisy reverberant environments to evaluate how well the proposed methods and a previous method improve signal to error ratio (SER) and Correlation between restored power envelopes. Results show that both proposed methods improve SER and Correlation more than the previous one.