Distortionless-Response vs. Matched-Filter-Array Processing for Adaptive Binaural Noise Reduction

Inhalt:
We present a binaural noise reduction algorithm based on the minimum variance distortionless response (MVDR) concept. This algorithm utilizes the relative impulse response between left and right ear which is estimated by means of the NLMS algorithm. In addition, the matched filter array (MFA) is analyzed as an approximation of the MVDR solution. The results show that the MFA performs more satisfactory than the MVDR at different input noise levels. Moreover, the impact of delay parameters in the system identification and in the filtering stage of the algorithms are investigated. Index Terms — Binaural signal processing, Minimum variance distortionless response, Matched filter array, Speech enhancement