Weakly Supervised Detection of Baby Cry

Abstract

Detection of baby cry is an important task of baby monitoring application. Effective and real-time detection of baby detection makes the baby well cared for while releasing the care giver’s pressure. Almost all existing methods for detection of baby cry use supervised support vector machines, CNN, or their varieties. In this work, we propose to use weakly supervised anomaly detection to detect baby cry in which baby cry is detected as an anomalous audio event. In this weak supervision framework, we only need weak annotation of if there is a cry in an audio file. We design a data mining technique using the pre-trained VGGish feature extractor and an anomaly detection network to obtain short audio files from long untrimmed audio files. The obtained dataset is 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 on untrimmed audio files or streams.

Received: 27 November 2023 | Revised: 18 February 2024 | Accepted: 10 May 2024

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in (1) GitHub at https://github.com/gveres/donateacry-corpus; (2) audioset at https://doi.org/10.1109/ICASSP.2017.7952261, reference number [11]; (3) ESC 50 at https://doi.org/10.1145/2733373.2806390, reference number [10].

Author Contribution Statement

Weijun Tan: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing - original draft, Writing - review & editing, Visualization, Supervision, Project administration. Qi Yao: Resources, Data curation. Jingfeng Liu: Resources, Data curation, Project administration.