A minimum-variance frequency-domain algorithm for binaural hearing aid processing
Michael E. Lockwood, Douglas L. Jones, Mark E. Elledge, Robert C. Bilger, Marc Goueygou, Charissa R. Lansing, Chen Liu, William D. O’Brien, Bruce C. Wheeler · The Journal of the Acoustical Society of America · 1999
A new algorithm has been developed that allows optimal filtering methods to be applied individually to different narrow frequency bands. Using this technique it is possible to process binaural signals in a manner which dramatically reduces the amplitude of signals originating away from a desired receive direction. The algorithm was tested on artificially combined anechoic and reverberant signals with varying numbers of interfering sound sources. In addition to being computationally efficient, the algorithm was able to produce output which had a consistently positive intelligibility weighted SNR gain [Link and Buckley, J. Acoust. Soc. Am. 91, 1662 (1992)], and in which the desired talker was noticeably easier to understand. Results of a paired-comparison listening test confirmed these results, and allowed an estimate to be made of the algorithm parameters which provided the best speech intelligibility for the output. These parameters included the window length, length of a filtered block, and amount of data which must be stored in memory. Collectively, the results show that this algorithm may be the first building block for a binaural hearing-aid processing algorithm which suppresses off-axis interference. [Research supported by the Beckman Institute.]