Pattern Recognition Modeling of Korean listeners’ perception of American English Monophthongs
Soonhyun Hong · 언어와 언어학 · 2015
This study investigated how Korean and American English listeners use what combinations of static and dynamic spectral properties, duration and F0 to identify American English monophthongs. A forced-choice identification test was administered to 57 Korean listeners and 5 American English listeners. Then a pattern recognition model (logistic regression classifier) was fitted to the results. It was found that Korean listeners use static spectral information for vowel identification while American English listeners dynamic information. Furthermore, Korean listeners use duration as a major cue for both high and non-high vowels while English listeners use duration for non-high vowels only.