Discrimination of Low-set Tongue Using Zero-crossing Rates and Mel Frequency Cepstrum Coefficients for Quantification of Pronunciation Speech
Yuka Ochiai, Kimiko Yamashita, Kaori Ishii, Masashi Nakayama, Shunsuke Ishimitsu, Satoshi Horihata, Kazutaka Kasai · International Journal of Oral-Medical Sciences · 2020
Oral habits, such as protruding and lowset tongue, cause malocclusion and may result in relapse after orthodontic treatment.Tongue position evaluation often relies on visual inspection; however, this approach does not provide objectivity and uniformity.Xray cinematography is the conventional method for evaluating tongue function, but exposure to radiation is inevitable.Since a palatogram and pressure sensor may affect physiological tongue movement, Acoustic analysis methods such as zerocrossing rates and Mel -Frequency Cepstrum Coefficients (MFCCs) were investigated to be an alternative.These metrics have been previously used, however it remains unclear whether it is possible to identify lowset tongue.In this study, we examined the discrimination of the lowset tongue during pronunciation.TPD5 is the distance between the center of the tongue and the palate.From our previous study, 16 participants with a TPD5 of 9.5 mm or more were classified as lowset tongue group by cephalometric analysis.9 participants with normal tongue positions were included in the normal tongue group.Audio from each group was recorded and the zero crossing rates and MFCCs were quantified.There were significant differences in MFCC2, MFCC4, and MFCC10 between the normal and lowset tongue groups.It was suggested that lowset tongue and normal tongue can be discriminated by MFCC2, MFCC4, and MFCC10.Keywords： low-set tongue, pronunciation speech, Mel-Frequency Cepstrum Coefficie nts, zero-crossing rates