P58 The influence of linguistic properties on recognition of naturalistic speech in noise
Speech in noise outcome measures are typically based on the recognition of linguistically controlled sentences presented with background noise. These measures often lack ecological validity because they fall short in accounting for perceptual and psycholinguistic variables inherent in real-life conversations. To understand how lexical properties of conversational speech material influence recognition of speech in noise, new stimuli were created based on the Diapix task [Hazan and Baker, 2011, JASA 130(4):2139; Van Engen et al., 2010, Lang. Speech 53(4):510], a dialogue elicitation paradigm that was developed to study talker- and listener-related adaptation strategies. Sixty-eight short sentences from the Diapix corpus were selected as test items, and recordings of the sentences were made with 8 talkers. The talkers were instructed to read the sentences clearly and naturally following the transcription, which has repairs and interruptions preserved. The talkers were also instructed to produce non-speech sounds (e.g., laughter, lip smacks, sighs). The continuous recordings were spliced manually to extract each sentence. The linguistic properties of interest include the number of content words in each sentence, overall sentence length, word position, phonological word length, lexical frequency, and phonological neighborhood density. Speech recognition accuracy in spectrally shaped steady-state noise with 2 dB SNR was measured using an online protocol with 31 young listeners. The listeners were native speakers of English and had normal hearing by self-report. Each listener heard speech from only one talker and was asked to provided transcriptions for each sentence. Linear mixed effect models were used to assess the effects of lexical properties on word- and sentence-level recognition accuracy in noise. Preliminary analysis suggests the effects of linguistic properties of the speech material on recognition accuracy in noise. The results demonstrate the impact of psycholinguistic properties on speech perception in addition to the perceptual difficulty introduced by noise. These findings can inform future studies aimed at developing clinical outcome measures that better represent everyday speech communication and more accurately capture listeners’ abilities in real-life scenarios.
Funding: Work supported by National Institutes of Health.