Comparing the consistency and distinctiveness of speech produced in quiet and in noise

The study investigated whether properties of speech produced in noise (Lombard speech) were more distributed (thus potentially more distinct) and/or more consistent than those from speech produced in quiet. This was examined for auditory tokens by measuring vowel space dispersion and by determining the consistency of formant production across repeated instances. Vowel space was not expanded for speech produced in noise; there was a tendency for formants to be produced more consistently in noise (with less variation in formant frequency across repeated instances) but this was not a secure effect. The distinctiveness and consistency of Lombard visual speech were also examined using motion capture data. Relative distinctiveness was determined by comparing the amount of mouth and jaw motion for speech produced in noise and quiet; relative consistency by comparing the size of correlations for motion produced across repeated instances in the noise or in quiet conditions. Mouth, and jaw motion was larger for speech in noise, however there was no greater association between the movement measures for repeated instances of speech in noise compared to in quiet. We also examined whether the correlation between auditory and motion properties was greater for speech produced in noise than in quiet. It was found that the association between speech RMS energy and jaw motion was greater for speech in noise. The results show that although Lombard speech affects both auditory and visible articulatory properties in ways likely to enhance speech perception it does not increase production consistency.