Korean oral stop recognition using durations

Weonhee Yun

Korean stops can be summarised as having three places of articulation and three manners of articulation for each place. Phonetic studies have reported that there is a durational difference between Tense vs. Lax stops, and Aspirated vs. Unaspirated ones. Tense stops have longer closure durations than Lax ones. VOTs of Aspirated stops are longer than those of their counterparts. These durational characteristics are statistically analysed in the KAIST speech database.

The KAIST speech database contains a variety of different sentences, which have influences on the durations of closures and VOTs, whereas the speech in the controlled speech data collected for phonetic studies is recorded in a frame or carrier sentence which prevents variation caused by linguistic factors other than the one being studied. However, even in the presence of contextual variability, the durational tendencies found in the KAIST speech database is still statistically consistent with the one in the controlled speech. Thus, closure durations and VOTs can be used as parameters for differentiating manners of articulation of oral stops.

On the basis of statistical study on the durations both in the KAIST speech database and in the controlled speech data, closure durations and VOTs for distinguishing manners of articulation of Korean oral stops are implemented in a Korean speech recognition system .