When Ariana Anderson ’05, Ph.D. ’09 had her first child shortly after completing her doctorate in statistics, she quickly discovered that her academic smarts were no help when it came to interpreting her baby’s cries. But by the time she had her third child, she had mastered baby speak.
“I could recognize patterns that led me to guess, at better than chance, what she needed — not because I had some superpower, but because I had already raised two other babies and there seemed to be a universal pattern,” says Anderson, an assistant professor of psychiatry and biobehavioral sciences in the Jane and Terry Semel Institute for Neuroscience and Human Behavior at UCLA. Ever the scientist, Anderson decided to test her hypothesis by seeing if she could build models to translate the baby sounds she had once struggled to understand.
The result is ChatterBaby, a free app that listens to a baby’s cry, assesses it in the context of a database that contains more than 20,000 baby sounds, then predicts the likelihood that the child is hungry, fussy or in pain. ChatterBaby relies on signal processing and machine learning algorithms to determine which sound frequencies and patterns are associated with which of the baby’s needs. The thousands of acoustic features analyzed include the strength, duration and amount of silence within the cries — for example, babies in pain have higher-pitched wailing periods with less silence, while fussy babies have lower-pitched crying periods with more silence. ChatterBaby was originally developed to help deaf parents understand the nature of cries they are unable to hear, but it soon became clear that all new parents could benefit. The app correctly flagged more than 90% of pain cries.
Anderson and her colleagues began developing the ChatterBaby database in 2013. They started by uploading some 2,000 audio samples of recorded infant pain cries, of which about 300 had happened during vaccinations or ear piercings. They enlisted a panel of veteran mothers to label other types of cries as hungry or fussy, using only unanimously agreed-upon sounds to teach the algorithm. Ever since the app went public in 2018, Anderson’s team has refined ChatterBaby with more user data — capturing not only the cries entered by parents, but also their conclusions about the source of the distress. As more data come in, the researchers continue to develop more sophisticated models, using algorithms powered by artificial intelligence to identify features that parents would otherwise be incapable of detecting.
While the parent input on why the baby is crying gets more specific (e.g. having colic, rash or fever or being tired) in the hope that ChatterBaby will at some point have sufficient data to increase its vocabulary, for now the focus is on what Anderson refers to as the three primal states that aren’t developmentally dependent — fussiness, hunger and pain.
“States like boredom, separation anxiety and fear don’t develop until the baby is 6 to 8 months old, and we wanted to make our app available to parents right from the get-go,” Anderson explains. She and her colleagues continue to test the reliability of parsing these categories, but in some cases that’s not feasible. For example, they’ve found that within the lower-pitched fussy-cry category, it’s impossible to distinguish between babies who are tired, have a fever or simply need a diaper change. Likewise, the cries of babies experiencing colic — defined as crying for three or more hours a day, three or more days a week for three or more weeks, and often associated with abdominal issues — sounded the same as those of babies experiencing more specific, short-term pain.
Based both on what she’s learned in developing ChatterBaby and her own maternal experience, Anderson offers the app’s users the following advice:
Calm Yourself First
“One of the most stressful parts of being a parent is when your baby is crying uncontrollably and you don’t know how to make them better,” says Anderson, now a mother of four. “Parents have a strong urge to make things good for their children, and when they’re not able to calm them down, they might see it as reflecting on their ability to take care of their baby, when it’s likely that they just haven’t tried the right thing.”
Trust Your Instincts
In assessing a baby’s cries, the ChatterBaby app offers only probabilities. Anderson emphasizes that it’s never meant to override the parent’s instincts. “Every baby is unique,” she says. “We did this to give advice, not prescriptions. It’s adding one piece of information that parents can use in putting together the puzzle.” As a new parent, she notes, it’s easy to second-guess oneself, but in fact, the parent’s intuition is correct much more often than many realize. “I always stress that the tools our algorithm [uses] are innate in parents,” Anderson says. “They are always going to be the best judges of what the baby needs.”
Act Quickly for Bonus Points
Not that any parent needs extra motivation to interpret and respond promptly and appropriately to their baby’s calls, but Anderson points to research showing that doing so contributes to language development. “When we try our best to identify and respond to the baby’s need, we’re showing them that their communication is being received,” Anderson says. “Otherwise, [for the baby] it would be like sending an email that keeps getting bounced back, which makes you not want to send it anymore.”
Choose the Right Sounds for Silence
When it comes to calming an inconsolable child, Anderson advocates swaddling — the age-old practice of wrapping the infant snugly with a blanket — in combination with soothing sounds. In particular, she recommends machines that emit brown and pink noise, which sound more like the ocean, rather than white-noise machines, which include higher frequencies.
When Nothing Works, Take Heart
Sometimes, as when a baby is experiencing colic, no parental response does the trick. Anderson offers empathy and counsels maintaining perspective. “It’s very hard when you have an inconsolable baby to not think you’re doing it wrong,” she says. “But the most easygoing of my four children was my hardest baby. A lot of these problems are self-correcting, and you just have to try your best, wait it out, and know that after the storm comes the sunshine.”
Read more from UCLA Magazine's April 2022 issue.