AI-Based Smart Baby Soothing System
Year: 2026
Type: preprint
Abstract: TwirlyTunes is an AI-based smart baby-soothing system designed to automatically detect infant crying and initiate soothing actions through lullaby playback and gentle jhoomar rotation. The system integrates edge artificial intelligence, audio processing, and embedded hardware to provide an automated infant-care assistance solution. The proposed system uses a Raspberry Pi 4, an INMP441 digital microphone, and a fine-tuned MobileNetV3-Small deep-learning model deployed using TensorFlow Lite. Captured audio is converted into log-Mel spectrograms and classified into Baby Cry and Non-Cry categories. A majority-voting mechanism is employed to reduce unnecessary activations. Once crying is confirmed, the Raspberry Pi initiates lullaby playback and communicates with an ESP32 microcontroller, which controls an L298N motor driver to rotate a jhoomar. The classification model achieved 95.06% accuracy, 94.42% precision, 76.89% recall, an F1-score of 84.76%, and a ROC-AUC of 0.9802 on a held-out test dataset of 1,479 audio samples. The project demonstrates the practical integration of lightweight deep learning, edge computing, and embedded systems for automated baby soothing. Further real-world testing is required to evaluate continuous detection reliability, response time, and mechanical safety. The system is intended as a supplementary infant-care assistance device and not as a replacement for adult supervision.
Institution: Somaiya Vidyavihar University
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7225431813