Shifting Sibilants: Spectral Sensitivity of the Lombard Effect in L1 English Speakers

Abstract

This paper examines the spectral sensitivity of the Lombard Effect in L1 English speakers with respect to sibilant consonants. Though previous literature has examined Lombard Speech produced under various noise conditions, no study to date has examined a ‘second-order’ Lombard effect by eliciting speech under speaker-specific noise designed to approximate a speaker’s typical Lombard Speech—that is, speech already produced under broadband noise conditions. Doing so elucidates possible different manifestations of the Lombard Effect and how such a noise condition might elicit speech which is differentiable from broadband Lombard Speech.

Results from spectral analysis indicate a weak-to-moderate sensitivity to the spectral characteristics of noise in the majority of speakers in this study, as demonstrated by adjustments to speech characteristics in both the temporal and frequency domains. Notable changes from broadband to speaker-specific noise include raising of peak frequency, increased segment duration, and decreased skewness. In conjunction, these changes do indeed differentiate speech between noise conditions in those speakers who first demonstrated acoustic changes from quiet to broadband noise conditions. Degree and manner of change in the frequency domain are highly speaker-specific, however, and occasionally segment-specific.

These findings may suggest that acoustic augmentations which differentiate the speech signal from tailored noise may be accompanied by articulatory changes meant to increase segment stability. Further work is needed to verify articulatory changes as well as the statistical significance of spectral trends for other measures across different noise conditions.