Fig 1.
(A) Overview of Mel-frequency cepstral coefficient (MFCC) extraction stages, including waveform segmentation, fast Fourier transform (FFT), and Mel-filter bank processing.
(B) Correlation analysis between MFCC layers using Discrete Wavelet Transform (DTW).
Fig 2.
(A) Fabricated palatal plate and (B) Experimental environment for data acquisition and photographs of the oral cavity wearing a fabricated palatal plate.
Table 1.
Target consonants of the speech materials used in the experiment.
Table 2.
Target vowels of speech materials used in the experiment.
Fig 3.
(A) Heatmaps visualizing the MFCC coefficient values for the consonants/tʃ/ and/k/, before (left) and after (right) wearing the palatal plate.
The horizontal axis represents time, and the vertical axis represents MFCC coefficient indices. The yellow dashed box highlights the specific coefficient range used in the correlation analysis. (B) Correlation coefficient matrix comparing MFCC patterns across different articulation conditions. Red indicates higher correlations (greater acoustic similarity), while green suggests lower correlations.
Fig 4.
(A) Spectrograms visualizing the formant structures and second formant (F2) slope for the consonants/tʃ/,/τ−Σ tΣ’/, and/ tΣh/, before (left) and after (right) wearing the palatal plate.
(B) Box plots comparing F2 slope comparisons across three articulation types: plain, tense, and aspirated. A nonparametric Friedman’s test revealed a significant main effect (p < .001), and post-hoc pairwise comparisons were conducted using the Bonferroni correction. (C) Box plots comparing F2 slope between normal and plate conditions for each articulation manner. No statistically significant difference was found based on the Wilcoxon signed-rank test (p > .05 for all comparisons).
Fig 5.
Comparison results of vowel formant frequencies (F1, F2, and F3) for normal and wearing the palatal plate.
(A) Line plots of average formant values for five vowels (/a/,/e/,/i/,/o/,/u/) under two conditions: normal test (solid lines) and wearing the plate (dashed lines). (B) Linear regression analysis showing the correlation between formant values under normal and plate-wearing conditions; and (C) box plots comparing the distribution of F1, F2, and F3 between the two conditions.
Fig 6.
Comparison results of the quadrilateral vowel space area (qVSA) before and after wearing the palatal plate.
The qVSA was calculated using the F1–F2 coordinates of the corner vowels/i/,/e/,/a/, and/u/. Blue and red shaded regions indicate the vowel space under normal (without plate) and wearing (with plate) conditions, respectively.
Fig 7.
Comparison of acoustic parameters during vowel production before and after wearing the palatal plate.