Apparent vocal tract length versus /s/ acoustics in a large multilingual corpus

The acoustics of /s/ are often studied for their socioindexical importance, especially as a cue to gender identity. It can, however, be difficult to distinguish socially meaningful variation from that which is introduced by anatomical differences between speakers—notably, by differences in vocal tract length (VTL). This differs from the situation for vowels, where formant frequencies scale predictably with VTL. A similar relationship might hold in sibilants, if front cavity length—and thus main spectral peak frequency—is roughly proportional to VTL. Alternatively, speakers may compensate for differences in VTL by subtly adjusting their articulation to achieve a particular acoustic target. To explore these possibilities, I examine the relationship between speakers’ apparent VTL (estimated from vowel formant measures) and the peak frequency of their /s/ productions in a multilingual acoustic dataset constructed from large read speech corpora. Preliminary results from nearly 1000 speakers across eight languages reveal that there is generally an inverse relationship between VTL and peak. Although its strength varies across languages, it appears stronger in Afrikaans, Czech, and English, more modest in French, Japanese, and Korean, and weaker or near zero in Mandarin and Arabic. This suggests VTL may influence the interpretation of patterns of interspeaker variation in sibilant acoustics.