윤태진. 2019. 이층위 신경망 모형을 사용한 포먼트 궤적 기반 모음 유형 분류. Studies in Phonetics, Phonology and Morphology 25.1. 95-112. 

The aim of this paper is to classify vowel types attested in the TIMIT corpus of American English using a two-layer neural net with the input feature of dynamic formant trajectories. Unlike previous studies which used a single (near the middle) or two static points (near the beginning and ending points within the duration of a vowel), the current study tried to incorporate feature values related to formant trajectories by sampling F1 and F2 values at every 10% of the duration across vowels produced in a connected speech corpus. Experimental results showed that the 20 vowel categories annotated in the TIMIT corpus could be accurately predicted at a rate of about 60.1% with a model of a two-layer neural net. The results confirmed that trajectory-based features were a better predictor then formant values obtained from one or two points. Even though the accuracy reported in the paper is comparable to or better than the accuracy rates reported in previous literature, further research needs to be done which can incorporate contextual information. (Sungshin Women’s University, Associate Professor)