Abstract:
Recent studies have demonstrated some success in improving speech intelligibility when either competing voices or reverberation are present, using an adaptive beamformer in the first case and a fixed-weight beamformer in the second. The present study evaluates the effectiveness of a hybrid beamformer incorporating the strengths of an adaptive beamformer with the advantages of an optimally chosen fixed-weight beamformer. Received signals at a short array of microphones in a room were simulated for a single talker and for a babble of voices at each of several interferer locations. Three rooms with different absorption coefficients were simulated. Conditions with more than one simultaneous interferer location and with starts and stops of an interferer are being tested. Success of the hybrid adaptive beamformer is being assessed by informal listening and by measuring the intelligibility-weighted improvement in SNR. Preliminary results appear promising.