Dynamic auditory-acoustic properties differentiate word-final fricatives in Gitksan

Gitksan is an endangered, understudied Tsimshian language spoken in northwestern British Columbia. This study aims to provide an acoustic-phonetic description of Gitksan fricatives. Isolated words containing word-final fricatives were recorded with three Gitksan first-language speakers (2 male, 1 female; age 65 + ). Postvocalic fricative tokens (s = 159, ɬ = 123, ç = 180, Xw = 150, χ = 267) were analyzed. Dynamic, spectral measurements of peak frequency in equivalent rectangular bandwidths (peakERB; Reidy, 2016) were estimated from 17 20-ms windows centered at equidistant points between the onset and offset of the fricative. The resulting trajectories (windows 3–15) show that the peakERB decreases as the place of articulation shifts further back in the vocal tract (alveolar s∼25–31, palatal ç∼20–23, uvular χ∼17–19; exception: labiovelar Xw∼15–16, likely lowered by labialization). Also, the lateral alveolar ɬ shows a lower peakERB (∼22–24) than the central alveolar s. A random forest model (Breiman, 2001) was trained on 586 tokens in predicting the fricatives with peakERB measures of the 17 windows. The trained model was tested on 293 tokens; it identified s most accurately and ɬ least accurately. Additionally, the model's relative-importance ranking of the 17 peakERB predictors suggests that the middle third of the peakERB trajectory is robust in differentiating word-final fricatives in Gitksan.