The new technique, which generated a significant boost in accuracy compared to the previously announced phone model, incorporates a data-driven cluster approach that utilizes k-means clustering to partition the acoustic feature space of child vocalizations. It has been known for many years that children with ASD have aberrations of voice and prosody. These differences between the vocalizations of typically developing children and children with ASD, though extremely difficult to identify with the human ear, can be identified statistically using advanced computer technology. The new technique was developed based on naturalistic full-day recordings from children diagnosed with ASD and children without ASD.
"Child vocalization decomposition could be done using either a phone model or clusters derived directly from child vocalizations," explained Dongxin Xu, Ph.D., manager of software and language engineering at the foundation. "The performances of the two methods are similar when applied individually. When combined together, the performance is significantly improved. This suggests that the two approaches capture different discriminant information for autism detection."
The LENA System comprises advanced processing software and specially designed children's clothing fitted with a lightweight LENA Digital Language Processor (DLP), a small, unobtrusive digital recorder. Designed for use in the natural home environment, the DLP can save up to 16 hours of high-quality audio, capturing all of a child's vocalizations as well as adult speech and other sounds.
About LENA Foundation
Established in 2009, the LENA Foundation develops advanced technology for the early screening, diagnosis, research, and treatment of language delays and diso
|SOURCE LENA Foundation|
Copyright©2009 PR Newswire.
All rights reserved