Integrative Modeling of Individual Differences Recognizing Speech in Noise by Hearing-Impaired Adults

Difficulty recognizing speech in noise is a common complaint among those with sensorineural hearing loss. Yet the degree of difficulty differs widely among individuals, often unrelated to the clinical gold standard for evaluating hearing, the pure-tone audiogram. Research has isolated both auditory and nonauditory factors responsible for these differences, but these factors do not operate in isolation. In the present work, a generic computational model involving simultaneous cue sensitivity, cue reliance, and decision noise provided an integrative framework for identifying sources of between-listener variance not accounted for by the audiogram. The framework was applied to performance differences within and between normal-hearing (NH) and hearing-impaired (HI) groups in the processing of linguistic, acoustic, and statistical cues supporting speech recognition in noise. The primary source of performance differences between groups was differences in sensitivity for the subtle, but largely stationary acoustic cues required for speech recognition. The overwhelming source of performance differences within groups was differences in decision noise associated with more salient, but highly variable statistical cues for speech separation. For speech separation, HI listeners placed far greater reliance than NH listeners on the one cue for which they were most sensitive. HI listeners, but not NH listeners, benefitted by shifting all acoustic information to this most relied on cue. The results provide preliminary support for the feasibility of integrative modeling as a means of evaluating the collective influence of factors affecting speech recognition in noise.