Speech enhancement based on aspects of the auditory process

Abstract

The effectiveness of a speech enhancement algorithm on both the intelligibility and quality of speech‐in‐noise by listeners with cochlear‐based hearing loss of moderate severity was tested. The enhancement scheme employed is based on an approach that uses the auditory masked threshold in conjunction with a version of spectral subtraction to adjust the parameters used in the subtraction process based on the masked threshold of the noise across the frequency spectrum. The ability of listeners with hearing loss to perceive stimuli from the nonsense syllable test (NST) and from the diagnostic rhyme test (DRT) in the presence of communication channel noise and automobile highway noise was measured at several different signal‐to‐noise ratios in both degraded and enhanced conditions. The algorithm effectiveness was also assessed through subjective ratings of speech quality. Results of the listening tests show that some but not all of the listeners with hearing loss benefited from the enhancement algorithm. We will discuss how the algorithm—which is based on models of auditory processing in normal‐hearing listeners—might be reformulated to address specific auditory deficits in listeners with cochlear hearing loss. [Work supported by National Organization of Hearing Research, Council on Research and Creative Work, University of Colorado.]