Effect of multiple nonstationary sources on MVDR beamformers
Michael E. Lockwood, Douglas L. Jones, Charissa R. Lansing, William D. O’Brien, Bruce C. Wheeler, A. S. Feng · 2004
MVDR beamformers have been proposed for use in hearing aids. Experiments show poor performance for time-domain methods for more speech sources than sensors, in contrast to the human auditory system. A frequency-domain binaural MVDR algorithm applied in a cocktail-party environment performs significantly better than the frost or GSC algorithms. Experiments show that the performance advantage of the frequency-domain algorithm can be attributed to rapid, independent adaptation across frequency bands, allowing faster, more accurate tracking of speech sources than time-domain techniques.