Study Finds LENA System Can't Track Speech Of Older Autistic Children
April 10th, 2019 By: JobsTherapy.com Content Staff
The LENA system, a popular tool for tracking speech development in autistic children, can’t distinguish the voices of children ages 5 and older from those of adults, suggesting that the system should only be used for younger children, according to a report by Spectrum, a leading source of news and expert opinion on autism research.
The Language Environment Analysis system features a wearable audio recorder and software that identifies the source of each recorded sound. Without it, researchers typically assess children’s speech by manually transcribing audio recordings, a labor-intensive process, Spectrum reported.
The LENA system was designed to track the voices of children ages 4 and younger and disregard the voices of adults and ambient sounds. The system monitors how often a child speaks and participates in conversations with an adult. A 2010 study found that LENA can tell the difference between the voices of autistic and non-autistic children ages 16 months to 4 years based on their speech patterns, sparking strong interest from researchers seeking to assess language therapies, Spectrum said.
“We were hearing a lot of buzz from people out in the clinical-trials world who would love the idea of using LENA because it essentially automates what can be a very laborious process,” Helen Tager-Flusberg, director of the Center for Autism Research Excellence at Boston University, who co-led the new study, told Spectrum.
“The pitch difference is very wide between a child’s voice vs. an adult’s voice, but as you get older, the pitch overlaps,” she added. “Then it becomes much harder for LENA to discriminate.”
Since no alternatives to LENA exist for older children, this study might lead to more research in this area, according to Jessica Dykstra Steinbrenner, advanced research scientist at the University of North Carolina at Chapel Hill, who didn’t participate in the study.
“Discovering that this particular tool doesn’t work for this age group is a good first step towards encouraging the field to develop something like this,” she told Spectrum.
Steven Warren, scientific adviser for the LENA Foundation, which makes the tool, told Spectrum that since LENA was created for kids ages 4 and younger, “I would have been surprised if it did work [for older children].” Warren is a distinguished professor of speech, language, and hearing at the University of Kansas.
According to the Spectrum report, Tager-Flusberg and other researchers analyzed LENA recordings of 51 children ages 5-18 and found that the software frequently mistakes the speech of older children for that of adults, missing more than half of the children’s utterances. Significantly, the older the child, the less accurate the software was. The study’s results were published in the Jan. 14 edition of the journal Autism Research.
“Make sure you don’t use LENA in kids 5 and up,” Rebecca Jones, assistant professor of neuroscience in psychiatry at Weill Cornell Medicine in New York, told Spectrum after participating in the study.
Jones and her research team are using transcripts of older children’s speech from the study to create a machine-learning algorithm that can distinguish between vocalizations of older children and adults, Spectrum reported. Their preliminary results suggest that the algorithm’s accuracy rate is higher than 80 percent.