Variation in the acoustic dynamics of /s/: Patterns within and across 32 languages

Sibilant fricatives have historically been considered fundamentally static speech sounds, characterized by clear articulatory and acoustic steady states. Recent work has called this idea into question, though, reporting differences in the articulatory and/or acoustic dynamics of sibilants across different sibilant sounds, languages, social groups, and phonological contexts. These results suggest that dynamics may be learned aspects of the production of sibilants, analagously as in vowels (including many monophthongs). However, previous studies differ methodologically in important ways, which may mean that at least part of the observed variation is driven by extraneous variables. A typological study where a consistent set of controls allows for direct comparison across a large number of languages and speakers is thus needed. In this paper, we examine the acoustic dynamics of /s/, the most common sibilant cross-linguistically, across 32 languages and more than 4,600 speakers using large read speech corpora. We find qualitative differences in the patterns of cross-linguistic and inter-speaker variation: whereas languages exhibit considerable variation in dynamics---particularly in the timing of the maximum of spectral center of gravity trajectories---most of the variation between speakers is instead in the overall height of the trajectory (i.e., in statics). We interpret this as evidence that sibilant dynamics are indeed learned, and discuss potential sources of the observed cross-linguistic differences.