Information Sci. Res. Lab., NTT Basic Research Laboratories, 3-1, Morinosato-Wakamiya, Atsugi-shi, Kanagawa, 243-01 Japan
This paper studies the variation of vowel and consonant articulation to examine the invariant features of phoneme-specific vocal tract configurations. The establishment of these features may be one of the goals of organizing articulatory movements, and they can be useful in constructing models of those movements. Articulatory data were gathered with a magnetic position-sensing device that traced the position of receiver coils placed on the lower jaw, upper lip, lower lip, and tongue in the midsagittal plane. Experiments were performed using (a) V[sub 1]V[sub 2]V[sub 3], (b) V[sub 1]V[sub 2]V[sub 3]V[sub 4]V[sub 5], (c) V[sub 1]CV[sub 2], and (d) CV[sub 1]CV[sub 2]CV[sub 3]CV[sub 4]CV[sub 5] contexts, where V[sub i] was a combination of vowels /a,i,u,e,o/ and C was one of consonants /p,t,k,b,d,g/. A Japanese male subject uttered these sequences at several speeds with different stress. Articulatory data were sampled when the vocal tract constricted for vowels or when the vocal tract closed for plosive consonants: For each vowel and consonant, 200 data frames were sampled on average from the whole data set of about 150 000 frames (10 min). The variability of the vocal tract configuration was then quantitatively analyzed for each phoneme by calculating the variation of each measurement point from its average position. The variation was very small when the vocal tract closed or constricted, especially in the traversal direction. How phoneme context, utterance speed, and stress affect the vocal tract configuration was also investigated.