Tongue Habit Discrimination System Using Acoustical Feature for Oral Habits Improvement

Abstract

Oral habits are tongue protrusion in malocclusions, causing deterioration of oral functions necessary for feeding, chewing, swallowing, and vocalization. In order to realize a non-invasive measurement of the habits, we propose and experiment acoustic feature analysis to discriminate tongue habits. Compared to normal speech, tongue-protruded speech is pronounced between the frontal teeth. Therefore, the speech is emphasized at a wide-range band of frequency components due to turbulence, as can be heard in the pronunciation of consonants. In this paper, we confirm these differences in acoustic features, such as zero-crossing which can capture the characteristics of voiced and unvoiced sounds and Mel Frequency Cepstrum Coefficient (MFCC) which is a filter bank analysis for front-end processing at speech recognition. We collect samples for that focus on the differences in oral habits of subjects, and significant of acoustic features which measured from the samples are confirmed. Finally, tongue habit discrimination using k-nearest neighbor algorithm (k-NN) achieved discrimination rate of about 85 to 98% on the databases.