Unpredictable and unintelligible: Individual differences in predictability-based reduction affect speech intelligibility

Abstract

Words spoken in high predictability (HP) contexts tend to be phonetically reduced and less intelligible than words spoken in lower predictability (LP) contexts. However, individual differences in degree and strategies of reduction, and their effects on intelligibility, are largely unexplored. This study examined the role of autistic traits in speech intelligibility. Sixteen participants completed a speech in noise word identification task at 2dB SNR. Stimuli were recordings of words spoken in HP and LP contexts, extracted from sentences produced by talkers ranging in autism-spectrum quotient (AQ) scores. After the identification task, listeners also completed the AQ questionnaire. A higher AQ score indicates a greater prevalence of autistic traits in one's cognitive style. Logistic mixed effect regression modeling revealed an expected effect of duration, such that longer words were more likely to be correctly identified. Further, talkers with higher AQ scores were less intelligible than talkers with lower AQ scores. Unexpectedly, no effect of predictability condition was observed: LP words were not consistently more intelligible than HP words. No effect of individual differences in listener AQ scores was observed. These results suggest that predictability-based enhancement and reduction strategies vary between individuals and are not necessarily for the benefit of the listener.