Acoustic surveillance system for children’s emotion detection

Abstract

This article presents a design of the system for acoustic event detection. The proposed pilot application is focused on child’s emotion/behaviour related sounds such as cry and laugh detection. Monitoring behaviour and safety of small children is particularly crucial in in-car environment and in households. The proposed application is based on Gaussian Mixture Model - Universal Background Model approach. The system is optimized by balancing false acceptance and false rejection rate. The classification accuracy of 71.6% was achieved although the system was trained only on small amount of data. The proposed approach has low computing and memory requirements thus is also suitable for implementation in embedded systems.