Predicting and classifying Estonian singleton and geminate consonants at various speaking rates using logarithmic duration

Amano et al. [2021, JASA, 150(3), 1830–1843] showed that a linear function using logarithmic durations can predict and classify Japanese singleton and geminate consonants at various speaking rates. However, it remains unclear whether this function can also be applied to other languages. To address this issue, Estonian singleton, geminate, and long-geminate consonants were analyzed using five minimal triplet words, spoken by four native Estonian speakers in carrier sentences at various speaking rates. Regression and discriminant analyses revealed that a linear function with two variables, the logarithmic duration of the consonant closure and the logarithmic average syllable duration of the carrier sentence, can predict the distributions of singleton (R2 = 0.57), geminate (R2 = 0.87), and long-geminate (R2 = 0.87). However, the singleton's function had a shallower slope than the others. The linear function can also classify the consonants with a small error rate (singleton versus geminate 9.6%; geminate versus long-geminate 9.2%). These findings suggest that the combination of logarithmic durations is an invariant acoustic feature of singleton and geminate consonants for speaking rate variation, and that this combination is a universal acoustic feature of the consonants across languages. [Work supported by JSPS KAKENHI Grant No. JP23K25325.]