Preliminary Statistical Pattern Recognition Methods in the Study of Vowels Produced by Children with and without Speech Sound Disorders

The nature of vowel acquisition and vowel error patterns in young children, especially those with speech sound disorders (SSD), is not well understood. Previous studies of vowels produced by children with SSD of unknown origin (SSD-UNK) have shown that they demonstrate a similar order of acquisition and vowel error patterns as those of children with TDS (typically developing speech), but are slower in developing vowels and produce more vowel errors (e.g., Stoel-Gammon & Herrington, 1990). Previous studies have relied mainly on phonetic transcription to describe the characteristics of vowels produced by children with SSD. A few studies used acoustic analysis to investigate the vowels of children with SSD due to motor impairments. These studies have shown that children with cerebral palsy have smaller vowel space size than children with TDS, and that size of vowel space is correlated positively with their speech intelligibility (e.g., Higgins & Hodge, 2001). These studies, however, are based on the acoustic measurements of only 3 to 4 corner vowels. In addition, despite the proposed importance of spectral movement patterns in successful identification of vowels (e.g., Hillenbrand, & Nearey, 1999), only a few studies (e.g., Lee, 2009) have examined spectral movement patterns of vowels produced by young children with SSD.