The Baby Cry AI

The Project

Our project is a web-based app called "AI Baby Cry Decoder." It uses machine learning to listen to a baby's cry and instantly predict why they are upset—whether they are hungry, sleepy, or in pain. It runs directly on any Android smartphone, turning the device into an instant AI translator for parents. ​What inspired it: We were inspired by seeing how stressful it is for new parents, especially young couples or couples with hearing difficulties, to understand a newborn baby's needs. Babies don't speak a language, but their cries have distinct acoustic patterns. We wanted to see if we could use deep learning to decipher this first language of humans and give parents immediate support. ​What was difficult: The hardest part was definitely the data and the deadline! We couldn't code an Android app from scratch in a weekend, so we had to find a creative workaround. Uploading hundreds of Kaggle audio files into standard tools kept failing. We had to learn how to use a professional platform called Edge Impulse, set up an "MFCC" algorithm to process human vocal frequencies, and train a Keras Neural Network. Testing the app on our phones and making sure it did not confuse normal background room noise with an actual baby cry was a huge challenge, but we finally got it working!

View this project on YouTube