5aSC5. Analysis of children's speech. Pitch and formant frequency.

Session: Friday Morning, June 20

Author: Sungbok Lee
Location: Central Inst. for the Deaf, St. Louis, MO 63110, lee@cidmv1.wustl.edu
Author: Alexandros Potamianos
Location: AT&T Labs, Rm. 2C-318, 600 Mountain Ave., Murray Hill, NJ 07974
Author: Shrikanth Narayanan
Location: AT&T Labs, Rm. 2C-318, 600 Mountain Ave., Murray Hill, NJ 07974


Pitch and formant measurements were performed on ten monophthongs of American English included in a speech database collected from 436 children (ages 5--18) and 56 adults [J. D. Miller et al., ICASSP, pp. 849--852 (1996)]. After automatic segmentation and labeling of each waveform, the pitch and the first four formant frequencies of each vocalic segment were estimated using a formant-tracking program (Entropic Research Laboratory, Inc.). Median values of each pitch and formant track were computed as the representative value of the track. The raw formant data were refined based on the linear regression statistics of each formant track and manual correction. The pitch and refined formant data were analyzed as a function of vowel, age, and gender. Results indicate that the rapid change in male pitch due to puberty occurs from age 12 to age 15, lower than the range (ages 13--17) in Kent [R. D. Kent, J. Speech Hear. Res. 19, 421--447 (1976)]. Age-dependent formant changes are explained by a linear scaling of the vocal tract length. The intra-subject variation of formant and pitch decreases as a function of age, supporting the results of a previous study [S. Eguchi and I. J. Hirsh, Acta Otolaryng., Suppl. 257 (1969)]. Implications of these and other findings on speech development will be discussed. [Work supported by AT&T.] XXSU SC

ASA 133rd meeting - Penn State, June 1997