1966 年 22 巻 1 号 p. 10-20
In case of vowels, vocal tract configurations or their equivalent area functions are described approximately by a parabolic model with three fundamental articulatory parameters: d_0 (a distance from glottis to the point of maximum constriction), r_0 (an effective radius of the maximum constriction) and g_L=A_L/l_L (a ratio of mouth-opening to lip-rounding). This model is modified to get closer description of Japanese vowel configuration drawn by Chiba and Kajiyama from X-ray photographs of the vocal tract configuration. The analysis of vowels by articulatory parameters can be carried out through the following two steps; (1) formant frequencies are extracted from observed frequency spectra through the analysis-by-synthesis technique, and (2) three fundamental articulatory parameters are extracted from those extracted formant frequencies by the table look-up technique with the consideration of the relations between three articulatory parameters and vocal tract configurations and further the relation between a vocal tract configuration and its resonance (formant) frequencies. The proportional variation of vocal tract dimensions under analysis is normalized and an efficient search on the table is also realized by the use of the ratio formant frequencies such as F_<21>=F_2/F_1 and F_<31>=F_3/F_1 instead of a direct use of these formant frequencies. A few results of articulatory analysis of vowels in connected speech are acceptable and giving quantitative supports to the qualitative description of articulatory dynamics of vowels found in the phonetics. The transformation of formant frequency representation of vowels into the articulatory-parameter-space seems to show a promising behaviour for separation and identification of each vowel.