Does vocabulary knowledge influence speech recognition in adverse listening conditions?
Degree GrantorUniversity of Canterbury
Degree NameMaster of Audiology
Purpose: To investigate the effects of vocabulary, working memory, age, semantic context, and signal-to-noise-ratio (SNR) on speech recognition in adverse conditions (multitalker babble) in normal-hearing listeners aged 18-35. First, a general hypothesis was tested that listeners with larger receptive vocabularies would be more accurate at recognising speech in noise than listeners with more limited receptive vocabularies, even when target stimuli are words with high lexical frequency. A second more specific hypothesis was that the vocabulary would be predictive of speech recognition accuracy when the signal was moderately degraded, but not mildly or severely degraded. Method: 80 sentences with a high (HP) or low (LP) degree of semantic predictability (40 HP and 40 LP) were recorded from a male speaker of NZ English. These sentences were used as experimental target stimuli, and presented in multitalker babble at four SNRs: -8, -4, 0 and 4 dB SNR. Thirty-five participants (11 males and 24 females, aged 18 to 35), with puretone hearing thresholds of 15 dB HL or better, completed the Peabody Picture Vocabulary Test (PPVT) and the Wechsler Adult Intelligence Scale (WAIS) vocabulary subtest, the WAIS working memory subtests, and the experimental listening task in which they were required to repeat back the target sentences. Results: There was considerable variability between listeners in speech recognition performance, in terms of percent words accurately recognised overall (M = 45.8%; SD = 7.4) and for both HP (M = 54.4%; SD = 9.8) and LP (M = 35%; SD = 8.9) conditions. Hierarchical multiple regression analyses revealed that receptive (PPVT) and productive (WAIS) vocabulary knowledge, but not working memory, contributed 8 significant variance to listeners’ speech recognition scores overall and in both the HP and LP conditions. Further regression analyses at individual SNR levels showed that receptive vocabulary contributed significant variance to listening recognition scores in all predictability and SNR conditions except the most favourable (HP stimuli at 4 dB SNR) and least favourable (LP stimuli at -8 dB SNR) listening conditions. Working memory and age were not significantly related to overall listening score, HP listening score, or LP listening score, but age did contribute significant variance in the - 4dB SNR LP condition. Conclusion: The results provide further evidence that greater vocabulary knowledge is associated with improved speech recognition in adverse conditions. This effect was salient in mid-range adverse listening conditions, but was not apparent in highly favourable and extremely poor listening conditions. The results were interpreted to suggest that in moderately adverse listening conditions listeners with larger lexicons may be better able to exploit redundancies and/or intelligible ‘glimpses’ in the speech signal.