Software Detects Autism in Toddlers' Speech In first test, it's correct 85% of the time By Nick McMaster, Newser Staff Posted Jul 21, 2010 3:20 PM CDT 3 comments Comments Leo Lytel, 9, poses for a photograph at his home in Washington Wednesday, May 6, 2009. Leo was diagnosed with autism as a toddler. He was undiagnosed at age 9. (AP Photo/Alex Brandon) (Newser) – The sounds made by a young child learning to speak can signal whether the child is autistic, a new study found. Researchers attempting a first pass at designing a voice-analysis software to identify autism found that their creation worked better than they could've imagined: the program identified autistic children correctly in 85% of cases, AOL News reports. Early detection is crucial because intensive therapy can start sooner. The voice analysis works by measuring the time children spend mispronouncing syllables as their acquire speech—autistic children tend to keep mispronouncing sounds longer. Currently, autism is diagnosed by a medical specialist who conducts a long observation period with the child, and the software, once finished, could be another screening tool. "We had no idea that this was possible," the lead researcher said. "It's very surprising that you can use a totally objective system and get this much information so quickly."