ASA 129th Meeting - Washington, DC - 1995 May 30 .. Jun 06
5pSC14. Time-variation in Mandarin voiceless sibilant spectra.
Yi Xu
Res. Lab. of Electron., Rm. 36-513, MIT, Cambridge, MA 02139
The three Mandarin sibilant fricatives, [s], [(hooked ess)], and [(looped
cee)], were studied using a combination of time averaging and ensemble
averaging as described by Xu and Wilde [ 3230 (A) (1994)]. Ten native
speakers of Beijing Mandarin (5 male and 5 female) produced these fricatives
followed by both unrounded and rounded high vowels. Time-averaged spectra were
obtained by averaging 13 consecutive 8-ms FFT windows over 20-ms intervals.
Token-ensemble-averaged spectra were then obtained by averaging over the
time-averaged spectra of 10 repetitions of each sentence produced by each
speaker at time-normalized locations in the utterances. Further
ensemble-averaging was also performed over the male and female speakers
separately to obtain group-ensemble-averaged spectra. Preliminary examination
of both token-ensemble-averaged and group-ensemble-averaged spectra reveals
that (a) Mandarin sibilants have substantial time variation in their noise
spectra, indicating continuous articulatory movement throughout the production
of the frication; and (b) all three sibilants manifest substantial spectral
variation due to anticipatory coarticulation with the following vowel. This
kind of variation is evident from the very beginning of the frication noise,
indicating that the extent of the anticipatory coarticulation may go beyond the
beginning of the frication. [Work supported by NIH.]