Evaluation of a dereverberation technique

Abstract

A two-microphone dereverberation technique was evaluated by obtaining speech recognition measures and preference judgments from normal-hearing and hearing-impaired listeners. Monaural speech recognition performance was measured for two reverberation conditions (0.4 second and 1.2 seconds) with and without processing. Binaural speech recognition performance was also measured for the unprocessed conditions. In addition, paired-comparison judgments of preference were obtained for all combinations of the processed and unprocessed monaural stimuli. For both groups of subjects, scores at the shorter reverberation time were significantly higher than scores for the longer reverberation time. For the normal-hearing subjects, processing to dereverberate had no significant effect on speech recognition performance. Binaural presentation of the unprocessed signal yielded significantly higher scores. For the hearing-impaired subjects, performance was significantly better in the unprocessed condition than the processed condition, but was not significantly different from the binaural condition. Paired-comparison judgments revealed differences in patterns of preference between normal-hearing and hearing-impaired listeners.