Young infants can now be screened for autism on the basis of the sounds they make. A recent American study suggests that an autistic infant’s babble differs from that of a normal child. Automated vocal analysis showed differences in 86% of cases, which opens up new possibilities for early diagnosis of autism.

Autism is a condition associated with anomalies in the development of the central nervous system, characterized by oral and/or non-verbal communication disorders, impaired social interaction and limited centres of interest. Over the last 10 years, the incidence of the condition has gone from one in 2,000 births to one in 150 births, which represents some 8,000 newborns a year, and the incidence of the disease continues to rise.

The researchers analysed the voices of 232 infants between the ages of 10 months and 4 years for a period of 1,500 days with a battery-powered all-day recorder snapped into their clothing. A total of over 3 million infant utterances were examined. The analysis focused on 12 specific acoustic parameters associated with speech development, and most notably syllabification, i.e. the infant’s ability to utter well-articulated syllables with rapid lip, jaw and tongue movement.

The researchers analysed each “speech-related vocal islands” (SVIs) while adjusting for differences in volubility across children and compared them to the expected results, which showed an absence of correlation in autistic children. Cases of autism have therefore been identified with remarkable accuracy (86%) thanks to an automated vocal analysis technique called Lena (Language Environment Analysis). This made it possible to differentiate between normal development, autism and language learning delay.

Thus, automated vocal analysis should in the near future help paediatricians screen for autism and thus assess the relevance of referring the child to a specialist capable of establishing a complete diagnosis and, if need be, suggesting a more effective treatment early in the child’s life. The authors acknowledge that this is their first attempt at automated vocal analysis and that the procedure can still be improved by including a greater number of children, considering other speech impediments along with different types of autism, and by taking other acoustic parameters into account. However, they are convinced that automated vocal analysis systems based on laboratory and naturalistically obtained recordings will prove highly valuable tools in identifying certain disorders, such as autism and learning delay.

Yet, because autism isn’t limited to language disorders (it also involves, among others, social skills), and because every infant develops at his or her own pace, such a method can never replace diagnosis by a specialist, but it is an early-screening tool well-worth considering.      

Source:
“Automated vocal analysis of naturalistic recordings from children with autism, language delay, and typical development" - Proceedings of the National Academy of Sciences of the United States of America (PNAS)