Infant Crying Detection in Real-World Environments

This paper addresses the problem of infant cry detection in real-world settings. While most existing cry detection models have been tested with data collected in controlled settings, the extent to which they generalize to noisy and lived environments, i.e., people's homes, is unclear. In this paper, we evaluated several established machine learning-based approaches as well as a promising modeling strategy leveraging both deep spectrum and acoustic features. This model was able to recognize crying events with F1 score 0.630 (Precision: 0.697, Recall: 0.567), showing improved external validity over existing methods at cry detection in everyday real-world settings. As part of our evaluation, we collected and annotated a novel dataset of infant crying compiled from over 780 hours of high-quality labeled real-world audio data, captured via recorders worn by infants in their homes, which we make publicly available. Our findings confirmed that a cry detection model trained on in-lab data underperforms when presented with real-world data (in-lab test F1: 0.656, real-world test F1: 0.243), highlighting the value of our new dataset and model.