Frequency and time domain auditory masking threshold constrained Kalman filter for speech enhancement

摘要

A constrained Kalman filtering algorithm based on auditory masking threshold is proposed for enhancing speech degraded by colored noise. The auditory masking threshold is used as a constraint to obtain a Kalman gain, which minimizes the estimate error variance under the constraint that the error power is smaller than the masking threshold. From the characteristics of the correlation vector, the power spectrum density and the masking threshold, a nonlinear constrained optimization problem is formed to calculate the Kalman gain. Simulation results show that the algorithm can improve subjective PESQ scores over both classic algorithms and recently published algorithms.