FUNDAMENTAL RESEARCH OF SPEECH DISCRIMINATION FOR TONGUE THRUST USING ZERO-CROSSING AND MELFREQUENCY CEPSTRUM COEFFICIENTS

摘要

This study investigates the differences in speech between the acoustic features of healthy people and patients of tongue thrust. Compared with healthy participants, patients of tongue thrust exhibit broadband spectral frequencies in speech accompanied by the non-stationary sound of a specific consonant. This paper proposes and tests two well-established acoustic features, zero-crossing and the mel-frequency cepstrum coefficients (MFCCs), to distinguish healthy speakers from those with tongue thrust. Zero-crossing here is the point where the frequency parameter changes sign, and is used to discriminate between voiced and unvoiced sections. The MFCCs represent cepstrum coefficients calculated on the mel scale of frequency followed by tone pitch. According to the results of tests of the proposed method on clinical datasets, significant differences between healthy participants and tongue thrust patients were observed in terms of both acoustic features, with p < 0.2 and p < 0.01, when conditioned on oral habits, gender, and age.