The playful babbling of young children may sound like gibberish, but those noises are the precursors to speech and tracking them may help identify autistic children -- perhaps at an earlier age than now possible, a new study suggests.

Researchers have developed the first computer program that can analyze hundreds of hours of voice recordings and accurately pick out children with abnormal or delayed speech patterns.

"This method could help track and predict mental development, and may even aid early detection of autism," said D. Kimbrough Oller, a professor of speech-language pathology at the University of Memphis and lead author of the study.

In the first years of life, infants typically go through several stages of speech development, initially experimenting with different pitches and sounds, including cries, squeals and babbles. Then they begin to integrate their tongue, jaw and lips to produce noises that sound like words, such as "bop bop bop." Shortly after, children learn to form simple words such as ball and duck.

"Children with autism, however, are almost always delayed in their language development," said Rhea Paul, director of the laboratory of developmental communication disorders at Yale University, who was not involved in the study.

Several years ago, Oller pinpointed 12 key characteristics of normal speech development, including the ability to control pitch and articulate syllables. However, at that time, there was no way to study these vocal features without manually transcribing and analyzing every hour of recording -- a laborious and time-consuming process.

That is why the LENA Foundation, a nonprofit focused on developing early screening technologies, created a novel computer program that can automatically analyze hours and hours of speech soundtracks using the 12 vocal parameters.

For the current study, published online July 20 in Proceedings of the National Academy of Sciences, Oller and his colleagues amassed almost 1,500 full days worth of recordings and identified 3.1 million sounds from three groups of children -- those with typically developing language skills, those with delayed skills, and those already diagnosed with autism.

Using the software, Oller's team was able to identify clear differences between the three groups. The typically developing children showed significant changes in their speech development over time, while autistic kids showed stunted growth. Language-delayed children fit somewhere in the middle, showing normal development for about half of the vocal features.

The researchers were also able to accurately predict the age of typically developing children based on their language development.

"This method is very powerful and the data are solid," said Paul. "The technology, however, cannot be used to replace the standard procedures for diagnosing autism." There are several key factors that contribute to an autism diagnosis, including difficulties interacting with others, which a voice recording would not pick up, he noted.

Oller suggests that this method could potentially detect abnormal speech patterns earlier in life. Typically, children are diagnosed with autism at 18 to 24 months, but if the technology is able to catch small nuances in speech patterns before 18 months, doctors could monitor these children more closely or potentially treat them earlier, Oller said.

However, Catherine Lord, director of the Autism and Communication Disorders Center at the University of Michigan, points out that scientists don't yet know all the right interventions to help children who are so young.

"It's too early to tell what the impact of this new method may be for identifying children suspected of having autism," said Lord, who was not involved in the study. "However, it's clear that studying these vocal differences between children with and without autism could lead to better interventions in the future."