How artificial intelligence may help decode your baby's cries — and contribute to autism research
Researchers hope their new Chatterbaby app could improve screening for autism
By her third baby, Ariana Anderson knew the difference between hunger cries and pain cries.
But that well-tuned ear didn't come quickly.
"Even though I'm a PhD, I'm a bit of a slow learner," the UCLA psychiatry and biobehavioural studies professor said.
"It actually took me three babies to realize what I feel like most parents realized by one baby."
Now, Anderson is making it easier for new parents to decipher their child's howls with an app called Chatterbaby that hopes to contribute to autism research.
Anderson and her team at UCLA collected recordings of crying babies, reported to be either hungry, fussy or pained. Using the app, a parent can record their baby's cries and an algorithm offers a prediction of how the child is feeling based on those three emotions.
"Within our own sample, it looks like it's about 90 per cent accuracy for flagging pain. About 70 per cent accurate overall," she told Day 6 host Brent Bambury.
The app's algorithm listens for a "few thousand" acoustic features — everything from the rhythm of a cry to its melody — to predict why a baby might be crying.
While the app has benefits for deaf and hard of hearing parents, Anderson is collecting recordings from parents in hopes of taking the research beyond hungry or hurt.
"Specifically, we want to assess whether or not babies who are at risk for autism have different cries early on," she said.
Over the past 15 years, researchers have been examining baby cries to determine differences in the way children with autism cry from a young age.
There is "lots of groundbreaking research that shows that babies who are at risk for autism show abnormalities in their cry at as early as six months of age," Anderson said.
"These abnormalities are so distinct that untrained listeners can listen to the babies cry and say, 'Oh that baby doesn't sound right,' at 18 months of age."
That's where Chatterbaby comes in. Anderson hopes that the app's algorithm will be able to pick up on those vocal abnormalities to help determine a diagnosis early on.
In the U.S., children who are white end up getting diagnosed almost two years earlier than children who are not.- Ariana Anderson
While the study has only recently launched, the app's availability allows Anderson and her co-researchers to reach parents they couldn't before.
"Usually people who participate in academic research studies happen to live close to the academic institution," she told Bambury.
"This means that people who live in other states, who live in other countries ... who have valuable information to contribute about their children … can't participate unless they are very close to us."
That's significant, according to Anderson. The age at which a child is diagnosed with autism depends on their income, she added,
"For example, in the U.S., children who are white end up getting diagnosed almost two years earlier than children who are not."
What's more, Chatterbaby offers parents with children 18 months or older behavioural assessments they can take to their doctor if they have concerns about their behaviour.
"What we want to do is to be able to empower parents everywhere if they have suspicions about their child's behaviour."
To hear the full interview with Marcus Gilmer, download our podcast or click the 'Listen' button at the top of this page.