Stability and Validity of Automated Vocal Analysis As a Measure of Vocalization Complexity in Preschoolers with ASD in Early Stages of Language Development

Background:

 

Explaining individual differences in spoken word use of preschoolers with ASD increases our understanding of variability in this population and helps us predict the extent to which children with ASD will use spoken words to communicate. Theory and research suggest that vocalization complexity explains individual differences in spoken word use in ASD.

Objectives:

This longitudinal correlational study examines the relative validity and reliability of three estimates of vocalization complexity in preschoolers with ASD who are preverbal or just beginning to use words to communicate. Two estimates were derived via automated vocal analysis of day-long samples of child vocalizations collected in natural settings, and another was derived from human coded, brief conventional communication samples collected in the lab.

 

Methods:

 

Participants were 33, 24-48 month old children with Autistic Disorder who were reported by their parents to use <200 words at entry to the study (Time 1). Our index of vocalization complexity from conventional communication samples was the proportion of communication acts including canonical syllables aggregated with the number of different consonants used communicatively across samples. The first automated index of child vocalization complexity, the infraphonological vocal complexity score, was derived using software developed for research by Oller et al. (2010) that can be applied after standard utterance labeling by LENA software. The additional automated index of vocalization complexity, the Automated Vocal Analysis (AVA) developmental age equivalency score, was provided directly by the standard LENA software. Parents reported children’s spoken vocabulary use on the MacArthur-Bates Communicative Development Inventories: Words and Gestures (MBCDI) checklist at Time 1 and four months later at Time 2.

Results:

 

The infraphonological vocal complexity score reached our criterion for acceptable stability with one day-long audio recording (g = .82) and covaried with Time 1 (r = .46) and Time 2 (r  = .51) spoken vocabulary. Associations for the infraphonological vocal complexity score with concurrent and future spoken vocabulary were non-significantly different from the analogous associations for the variable from conventional communication samples (Z = -1.12, p = .26; Z = -1.02, p = .31, respectively). The AVA developmental age equivalency score was similarly stable, but was not significantly correlated with concurrent or future spoken vocabulary in our sample. Results were similar for the subset of our sample reported to use <20 words on the MBCDI at Time 1.

Conclusions:

 

Results suggest the infraphonological vocal complexity score from automated vocal analysis is a valid and reliable alternative to the more expensive vocal complexity measures from conventional communication samples and may thus provide a cost-effective method for measurement of vocal complexity in clinical practice. However, at present the infraphonological vocal complexity score is not publicly available in the standard LENA software package.