Modeling retroflex fricative variation in accented mandarin

Since dental-retroflex fricative contrast is not consistently maintained in many southern dialects of Chinese, native speakers of these dialects may not accurately produce the Mandarin retroflex fricative /ʂ/. Consequently, /ʂa/ may be realized as [sa] (Duanmu, 2007). This study investigated the variation of the retroflex fricative /ʂ/ in a Chinese Mandarin speech corpus (DataTang, 2018). The corpus contains 200 hours of recordings of 600 speakers from different dialectal regions in China. Each recording was aligned at the phone level using Montreal Forced Aligner. The center of gravity of the acoustic energy (COG) of the target sounds was extracted using Christian DiCanio’s Praat script. For statistical analysis, the generalized additive mixed-effects model (GAMM) was used. COG was the response variable. The following vowel’s height, tone, and gender were factorial predictors. To evaluate the geographic effect, we used tensor product smooths by fricatives with the longitude and lat...