The purpose of this study was to identify the main factors that differentiate listeners with clinically normal or “near-normal” hearing with regard to their speech-in-noise perception and to develop a regression model to predict speech-in-noise difficulties in this population. We also aimed to assess the potential effectiveness of the formula produced by the regression model as a “diagnostic criterion” for clinical use.
Data from a large-scale behavioral study investigating the relationship between noise exposure and auditory processing in 122 adults (30 to 57 years) was re-examined. For each participant, a composite speech-in-noise score (CSS) was calculated based on scores from three speech-in-noise measures, (a) the Speech, Spatial and Qualities of Hearing scale (average of speech items); (b) the Listening in Spatialized Noise Sentences test (high-cue condition); and (c) the National Acoustic Laboratories Dynamic Conversations Test. Two subgroups were created based on the CSS, each comprising 30 participants: those with the lowest scores and those with the highest scores. These two groups were compared for differences in hearing thresholds, temporal perception, noise exposure, attention, and working memory. They differed significantly on age, low-, high-, and extended high-frequency (EHF) hearing level, sensitivity to temporal fine structure and amplitude modulation, linguistic closure skills, attention, and working memory. A multiple linear regression model was fit with these nine variables as predictors to determine their relative effect on the CSS. The two significant predictors, EHF hearing and working memory, from this regression were then used to fit a second smaller regression model. The resulting regression formula was assessed for its usefulness as a “diagnostic criterion” for predicting speech-in-noise difficulties using Monte Carlo cross-validation (root mean square error and area under the receiver operating characteristics curve methods) in the complete data set.
EHF hearing thresholds (p = 0.01) and working memory scores (p < 0.001) were significant predictors of the CSS and the regression model accounted for 41% of the total variance [R2 = 0.41, F(9,112) = 7.57, p < 0.001]. The overall accuracy of the diagnostic criterion for predicting the CSS and for identifying “low” CSS performance, using these two factors, was reasonable (area under the receiver operating characteristics curve = 0.76; root mean square error = 0.60).
These findings suggest that both peripheral (auditory) and central (cognitive) factors contribute to the speech-in-noise difficulties reported by normal hearing adults in their mid-adult years. The demonstrated utility of the diagnostic criterion proposed here suggests that audiologists should include assessment of EHF hearing and working memory as part of routine clinical practice with this population. The “diagnostic criterion” we developed based on these two factors could form the basis of future clinical tests and rehabilitation tools and be used in evidence-based counseling for normal hearers who present with unexplained communication difficulties in noise.