| Abstract [eng] |
The analysis of spectral formant movement of Lithuanian vowels was carried out. More frequently used words with different surrounding of sounds “a“, “e“, ”ė”,” i”, “o”, “u” were selected for investigation. Praat program was used for words annotation into phonemes and obtaining values of formants. The movement of formants was approximated by other sound parameters, as duration, time moment from sound pronunciation beginning, intensity and its polynomial approximation coefficients. Approximation was carried out, using artificial neural network. The result was more narrow remaining errors distribution. The sounds were divided into groups by their stress, because parameters of approximation have less spread in these groups. The formant movement dependencies versus time were obtained. The generalized frequencies for formants F1 and F2 were calculated, when time is in the middle of the sound and other parameters have zero values. |
