Skip to main content

To a trained ear, a child's early speech patterns can hint at autism. But as a screening method, detecting speech and voice problems can be difficult, and is not widely used.

Now, new research suggests that an automated system may one day take some of the guesswork out of detecting the neural disorder in young children by recognizing deviations from the norm in voice recordings

Lead author Kimbrough Oller and his colleagues analyzed nearly 1,500 full-day soundtracks from 232 children, ages 10 months to four years. The children wore recording devices snapped onto their clothing. About half of the children were considered to be typically developing. Of the remaining children, about 50 had been diagnosed as language-delayed and about 75 had been diagnosed with autism.

"There's a robust difference between the groups," says Dr. Oller. The study was published Monday in the U.S. journal Proceedings of the National Academy of Sciences.

The automated system separated the sounds made by the children from background noise and then classified each utterance according to common characteristics of vocal development, including creating syllables, articulation and pitch.

The program was able to correctly classify the typically developing children by their age. It was then able to differentiate the autistic kids from the typically developing kids 86 per cent of the time, says Dr. Oller, a professor at the School of Audiology and Speech Language Pathology at the University of Memphis.

When looking at these two groups plus the language-delayed kids, the software worked 79 per cent of the time in tracking autism - mostly because autistic children and children with language delay can have similar, overlapping deficits.

The deficits in the speech and voice patterns of the autistic children are particularly difficult to quantify because they can appear in various combinations and in varying degrees over time.

But as an example, Dr. Oller says that autistic children appear to be very slow in developing well-formed syllables, such as "ba." Deviations might include saying "ba" with a tremor or "ba" with a trailing "aaah" sound. These are problems that speech experts would identify as having to do with voicing, slow transitions or simply the duration of the sound.

Other measures are likely to remain stronger indicators than speech in screening for autism, Dr. Oller says. One such measure is known as "joint attention," in which infants point at objects then look to a parent or caregiver to, in effect, share the thought. This usually happens around one year of age, but autistic children don't do it or do much less of it than typically developing children.

But he says his research proves that speech and voice do have a place in the diagnosis of autism, he says.

"People have always claimed that vocalization is different (with autistic children). There is now a way to [detect]it and it can be done totally objectively."

Interact with The Globe