Predicting and classifying Japanese singleton and geminate consonants using logarithmic duration.

To clarify the acoustic variables for predicting and classifying Japanese singleton and geminate consonants, raw and logarithmic durations of the consonants and their related segments were examined using 12 minimal pair words that were pronounced in a carrier sentence at various speaking rates by 20 native Japanese speakers. Regression and discriminant analyses revealed that the logarithmic durations were better at predicting and classifying Japanese singleton and geminate consonants than the raw durations used in many previous studies. Specifically, the best acoustic variables were the logarithmic duration of the consonant's closure or frication and the logarithmic average duration of the mora in the preceding carrier phrase. These results suggest that logarithmic durations are relational invariant acoustic variables that can cope with the durational variations of singleton and geminate consonants in a wide range of speaking rates.