Robust Broadband Beamformer with Diagonally Loaded Constraint Matrix and Its Application to Speech Recognition

In this paper, we develop a robust wideband adaptive beamforming algorithm for microphone array based speech recognition. We develop a quadratic constraint based approach to deal with the uncertainty in the look direction. In order to address the ill-conditioning associated with the constraint matrix, diagonal loading (DL) is employed. The advantage of adding DL to the constraint matrix is that the constraint matrix is only determined by the geometry of the array thereby allowing the DL level to be chosen offline. We also develop an iterative algorithm (and corresponding adaptive algorithm) to solve for the robust beamformer coefficients. The developed algorithm is applied to the problem of beamforming using microphone arrays for speech recognition and shown to be superior to existing algorithms.