Vowel data of early speech development in several languages

It is notoriously difficult to perform reliable spectro-temporal analyses on speech of young children (up to two years of age), partly because of the high pitch of their voices. Formants are very poorly defined and thus we used a pitch-synchronous bandfilter analysis, followed by a principal components analysis to represent the acoustic characteristics of children’s vowel(-like) productions. Since young children’s vowel realizations are very hard to label consistently, it is fortunate that this analysis method works just as well for unlabeled data. This way we can still study the size and form of the acoustic vowel space and its changes over the first two years of life. This is indispensable information to develop a universal vowel acquisition theory and to establish effective treatment methods to remediate disordered vowel productions. So far we have analyzed vowel data from 5 to 8 boys at two years of age in each of the following three languages: Dutch, Hungarian and Russian. Many practical difficulties were encountered (related among other things with the ways of collecting and actually recording the vocalizations), that made comparisons within and between languages not always easy, but nevertheless some language specific properties appear to be detectable. 1