Developmental vowel acquisition data obtained by a band-filtering analysis method: Results from Hungarian children

Abstract

A recently developed band-filtering analysis method (van der Stelt et al., 2003, 2005; Zajdó et al., 2004) that minimizes the dependence of the results on fundamental frequency allows for a dependable identification of spectral measures in children’s vowels. In the scripts, selection of segments with certain acoustic features (e.g., sound quality, F0) is followed by the calculation of spectral envelopes between 0–7 kHz, resulting in 40 filter values per spectrum. An intensity matrix is produced with data from each 40 filters. Data reduction is achieved through prinicipal component analysis. Results indicate that the first two principal components are related to the F1 and F2 of vowels, both in adults (Pols, 1977) and children (Zajdó et al., 2005). Analyses of unlabeled vowel productions in Hungarian-speaking children at 2; 0, 3; 0 and 4; 0 years show that the first two eigenvectors account for >50% of the variability in all age groups. By using these eigenvectors, age-specific reference planes are created. Mapping phonemically labeled vowel measures onto the 4; 0-year-old’s plane from each age group suggests that children produce decreasingly overlapping vowel categories as they age. With development, vowel categories are increasingly specified, covering more distinct areas of the acoustic vowel space. [Work supported by UW.]