Children with autism may in future be diagnosed by the way they talk and parents may be able to measure their progress by monitoring their speech. Scientists have discovered that the condition has a "unique vocal signature" which could allow affected children to be identified before they show obvious symptoms.

Swift diagnosis would allow parents to organise remedial help, which has the best chance of working if started early. Researchers say autism could be diagnosed as early as 18 months compared with the current average age at diagnosis (in the US) of 5.7 years.

The new method involves recording speech sounds made by very young children and analysing them with a computer programmed to recognise distinctive patterns. Scientists tested the method on 232 children aged from 10 months to four years previously diagnosed with autism or language delay. They were able to identify those with autism with 86 per cent accuracy.

The children were given battery-operated tape recorders, which they carried all day. Almost 1,500 recordings were collected and analysed, and the results compared with those for children developing at a normal pace.

The biggest difference was in the ability of the children to form syllables. This requires rapid movement of the jaw and tongue, a skill infants normally acquire in the first months of life and later refine as they learn language.

Children with autism were much slower to develop "syllabification" than others of the same age. Even those with delayed language development progressed faster. Steven Warren, professor of applied behavioural science at the University of Kansas, who was involved in the research said scientists had suspected autism affected the way children talked.

He and his colleagues developed the computerised system which for the first time allows the collection and analysis of huge quantities of data at relatively low cost. "This technology could help paediatricians to screen children for autism spectrum disorders to determine if a referral to a specialist for a full diagnosis is required and get those children into earlier and more effective treatments," he said.

The analysis is not based on words but on sound patterns, so the technique could theoretically be used with children speaking any language.

"The physics of human speech are the same in all people as far as we know," Professor Warren said, adding: "Autism interventions remain expensive and arduous. This tool may help us to develop cost-effective treatments and better understand how they work and how to keep them working."

Parents who provided language therapy for their children could assess its effectiveness for themselves at home, using the technique.

The findings are published in Proceedings of the National Academy of Sciences.