At Bayes Studio, we have always believed that innovation is most valuable when knowledge is shared. Alongside developing practical AI, sensing, and robotics technologies, we are committed to contributing to the broader knowledge base through research, field studies, technical publications, and collaboration with academic, industry, government, and community partners. By openly sharing our findings, lessons, and perspectives, we aim to support informed discussion, accelerate responsible innovation, and help advance solutions to complex environmental, public-safety, and security challenges.
Our Research:
Integrated Acoustic Monitoring & Prediction System: Lightweight Edge-Deployable Environmental Sound Classification
This work presents an edge-based acoustic monitoring system for real-time thunder and fire detection in mountain environments. Using 4,041 clips across six sound classes, it evaluates multiple feature extraction methods and three lightweight models. A BC-ResNet with 5-second tiled mel spectrograms achieved 82.6% accuracy, 88% thunder recall, and ~50 ms inference on a Raspberry Pi 4.