The influence of semantic information on the acoustics of speech in noise

Abstract

While there is a significant body of work on how people modify their speech patterns in the presence of noise, the role of semantic information on these acoustic modifications is not well understood. This study examined whether adult speakers of English differentially modify semantically salient versus nonsalient words within a sentence in the presence of noise. Participants were asked to produce a set of 20 sentences [from the Speech Perception in Noise List, Kalikow et al. (1977)], in each of five noise conditions; quiet, 60 and 90 dB SPL multispeaker conversation, and 60 and 90 dB SPL street noise. Five random repetitions per sentence were requested for each noise condition. The following acoustic cues were extracted for each word within an utterance: duration, average intensity, peak intensity, average fundamental frequency, and peak fundamental frequency. The ratios of these measures were then compared for semantically salient versus nonsalient words. While we expect to see an overall change in all of these prosodic cues for speech produced in noise (Lombard effect), semantically salient words may exhibit a greater ratio of change in some or all features.