As any parent will tell you, it’s almost impossible to interpret a baby’s cry.
The Infant Cries Translator app promises to decode whether your little one is hungry, tired, in pain or has a wet diaper by comparing it to a database of 200,000 cries from 100,000 newborns.
The app, which was developed by researchers at the National Taiwan University Hospital, Yunlun, takes just 15 seconds to analyze a baby’s cry, before sending a message (translation) to a parent's smartphone. Lead researcher Chang Chuan-yu explains how this service can help any concerned parent.
Q: What led you to the development of the Infant Cries Translator?
Babies are born with the ability to cry; it is how they instinctively communicate with their parents. According to reports of pediatricians, normal newborns cry two hours a day. However, it is sometimes difficult for parents to figure out why the baby cries. Our translator recognizes different types of infant crying to help parents and nursing staff take care of infants.
Q: What was the process to create this app?
We have collected more than 200,000 cries from the National Taiwan University Hospital Yunlin Branch. Machine learning and the big data analysis techniques are used to automatically adapt the crying model for each infant. Infant Crying Translator recognizes four types of infant crying, including hunger, pain, wet-diaper and sleepiness.
Q: How does it work?
Infant Crying Translator is the first commercial app to have the ability to learn the newborn cries. It provides parents of newborns the chance to record a baby's cry and create a unique baby crying model for your baby. This will increase the recognition rate dramatically. Just push the record button when the baby is crying and the app will upload the cries to EmoRec Cloud. Some acoustic features of the cry signal will be extracted and compared to the four crying models. The classification result will the be sent to the user’s phone. In case the classification result is incorrect, the user can annotate the true reason for the crying, then the cloud will re-train the crying model for the baby.
Q: How accurate is this app?
Experimental results have revealed a good performance: the recognition accuracy is up to 92 percent, 85.4 percent, 83.8 percent, and 77 percent for infants born within two weeks, one month, two months, and four months, respectively.
Q: Why doesn’t this app work well after six months of age?
Because babies born after six months have started to learn to speak, have their own emotions, and parents can communicate with their baby.
Q: What is your next project?
A: Since there are various phones with different Android versions, we will try to make the app compatible with most phones. We will also continue to enhance the recognition rate of the app. In the near feature, we will launch a new product, the core function of the “Infant Crying Translator" will be built-in a surveillance camera.