Weakly Supervised Detection of Baby Cry

Abstract

Detection of baby cry is an important part of baby monitoring. Almost all existing methods use supervised SVM, CNN, or their varieties. In this work, we propose to use weakly supervised anomaly detection to detect baby cry. In this weak supervision framework, we only need weak annotation if there is a cry in an audiofile. We design a data mining technique using the pre-trained VGGish feature extractor and an anomaly detection network on long untrimmed audiofiles. The obtained datasets are used to train a delicately designed super lightweight CNN for cry/non-cry classification. This CNN is then used as a feature extractor in an anomaly detection framework to achieve better cry detection performance.