Analysis and synthesis of children's speech Analysis and synthesis of children's speech
To anyone who has had a conversation with a 3-year old, it is apparent that the speech of young children is inherently more difficult to understand than that of older children or adults. Children are harder to understand for a variety of reasons: limited memory span, problems with attention, and the fact that their cognitive and linguistic systems are still developing. One of the most important reasons, however, is that young children's speech is quite different and much more variable than that of older children and adults. This variability has been a major impediment for clinicians attempting to diagnose and remediate childrens' speech disorders; for theorists concerned with the link between speech production and perception; and for engineers developing systems of computer speech processing. While there have been numerous studies of children's speech, the majority have been isolated kinematic, acoustic or perceptual investigations of specific speech features. Few have brought together these three types of data with the aim of explaining variability in different age groups. This research takes three important steps to overcome this limitation: First, acoustic and kinematic data are collected simultaneously, using magnetometer technology; second, acoustic and kinematic analyses are carried out in parallel with listening experiments to evaluate the perceptual consequences of acoustic and articulatory variation; third, statistical pattern recognition models will be developed to characterize the mapping between articulation, acoustics, and perception. This research has two additional practical objectives: (i) to evaluate the reliability of formant-tracking algorithms for analyzing children's speech; (ii) to evaluate the intelligibility of synthesized children's speech generated with different synthesis algorithms.
Research in progress: we are assembling a large database of CVC syllables from 3, 5, and 7-year old children as well as adults. An initial set of recordings of /hVd/ words have been used in a series of perceptual experiments that include natural and synthesized children's vowels as stimuli.
Assmann, P.F., Katz, W.F., Jenouri, K.M. & Hamilton, P.R. (1995). Identification of natural and synthesized vowels produced by children and adults: Effects of formant frequency variation. Journal of the Acoustical Society of America 98, Suppl. 2, 4pSC3, p. 2964.
Katz, W.F., Assmann, P.F., & Jenouri, K.M. (1995). Identification of natural and synthesized vowels produced by children and adults: Effects of fundamental frequency variation. Journal of the Acoustical Society of America 98, Suppl. 2, 4pSC2, p. 2964.
Research grant: Texas Advanced Research Program Grant # 11423-590, Awarded by the Texas Higher Education Coordinating Board, "Acoustic, kinematic and perceptual studies of children's speech" (William Katz, principal investigator and Peter Assmann, co-principal investigator).