gms | German Medical Science

The effect of tonsillectomy on the voice and the articulation were performed in several experimental studies. Hori et al. [Ref. 1] studied the formant frequencies changes of 5 Japanese vowels. It was found that only the third formant (F3) decreased remarkably especially for vowel /O/. Saida et al. [Ref. 2] analysed Japanese vowels and Japanese conversational sentences. Although there was a tendency for a decrease in formant F3, tonsillectomy did not appear to change the acoustic features remarkably. Ilk et al. [Ref. 3] evaluated the center formant frequency of Turkish vowels and MDVP parameters for perturbance measurement (jitter, shimmer, noise-to-harmonic ratio). The changes were found at the position of formant F3 and its formant bandwidth B3 for the vowel /O/ and slight decrease in B2 and B3 for the vowel /A/. The noise-to-harmonic ratio also decreased slightly, suggesting less nasalized vowels.

The biggest disadvantage of such studies is that the patients are not able to repeat the same manner of voice production before and after tonsillectomy. The results can be evaluated statistically only. The advantages of computational modelling methods are obvious, because they could give more precise theoretical results nowadays.

This paper presents preliminary results of mathematical modelling of acoustic characteristics and phonation for vowels /A/ and /I/. The results were compared with the acoustic voice analysis of patients before and after tonsillectomy.

After the phoniatric examination the tape-recording of voice was carried out on 14 patients (6 male, 8 female, age from 16 to 39 years) in a studio and digital photographs from mouth cavity were obtained before and after the tonsillectomy. The second recording was performed approximately 1 month after tonsillectomy. The age of patient were chosen to correspond to the computational model. They phonated five Czech vowels /A/, /E/, /I/, /O/, /U/. The acoustic analysis was performed by Multi-Dimensional Voice Program (MDVP Advanced) and the position of the first four formants was evaluated. The statistical analysis was evaluated by nonparametric statistics for two groups (Wilcoxon). The size of tonsils was subjectively estimated from photograps in the scale: 1 - small tonsils (5 patients), 2 - middle tonsils (6 patients), 3 - large tonsils (3 patients).

The finite element (FE) models of the acoustic spaces corresponding to the human vocal tract for the Czech vowels /A/ and /I/ were used in the mathematical modelling [Ref. 4], [Ref. 5]. Two FE models were created for each vowel for two different sizes of the tonsils, the first model with the volume 1.6 cm³ per tonsil and the second model with the volume of 0.8 cm³ per tonsil. The acoustic resonant characteristics of the FE models were studied by modal and transient analyses (using excitation by a short pulse). Transient analysis was also used for numerical simulations of the vowels production. The FE models of the vocal tract of a normal male subject were created using magnetic resonance imaging (MRI) technique and the FE mesh of a hollow spherical space, representing the acoustic space around the human head, was added manually (Figure 1 [Fig. 1]). The acoustic analyses were realised by the software code SYSNOISE.

Statistically significant changes of formant positions at 1% level were found at the experimental study for formant F3 for the vowels: /A/ (average decreasing about 200Hz), /E/ (decreasing about 150 Hz) and /O/ (decreasing about 240 Hz). The changes of formant position are dependent on the size of tonsils (Table 1 [Tab. 1]). The largest shift was found for vowel /O/ as the other studies reported. The perceptual differences by subjective assessment are noticeable only if the two samples are following immediately one after the other (Figure 2 [Fig. 2]).

The formants (eigenfrequencies) computed by the modal analysis of the vocal tract alone are practically the same as resonant frequencies computed by transient analysis using the FE model with the spheres respecting radiation of the sound into an open space.

According to the results of the modelling for the vowel /A/, the tonsillectomy caused the shift of the formants F3 and F4 down to the lower frequencies of about 180 Hz and 118 Hz, respectively. The other formants were practically unchanged. For the vowel /I/, the tonsillectomy caused a frequency shift down of about 100 Hz for the formants F2, F4, and F5. For the model with the smaller tonsils the frequency shifts of the formants were approximately two times lower.

The statistical analysis of experimental results of formant frequency positions before operation and one month after operation proved that changes were significant at the formants F3 of vowels /A/, /E/ and /O/. The decrease of formant F3 was reported in literature by the other authors [Ref. 1], [Ref. 2], [Ref. 3]. The largest decrease was found for vowel /O/. This finding also corresponds to the results which can be found in other languages.

The computed eigenfrequencies for FE models are similar to the experimental data known for formants of the Czech vowels /A/ and /I/ from literature [Ref. 6], and the resulting formant changes due to the tonsillectomy are in a reasonable agreement with the experimental data as well.

This research is supported by the Grant Agency of the Czech Republic by project No 106/04/1025 "Modelling of vibroacoustic systems focusing on human vocal tract".