Investigation Of Infants’ Crying Detection In Noisy Home Scene With Deep Learning

Abstract

In research of rich media data, how to extract and utilize semantic information based on the data is essential. With the development of deep learning technology, semantic information extraction and analysis is becoming more mature. Regarding the recent research on audio data in a home scene, infants’ crying is a specific topic besides human speech sounds. However, in current research, infants’ crying data are usually generated from quiet laboratory environments with limited noise, and there are few studies on the detection of infants’ crying in a high-noise environment, while noise is inevitable in the home scene. Based on this observation, this paper will study the feasibility of using deep learning techniques to achieve crying detection in high-noise environments. The results have identified that the Convolutional Neural Networks (CNN) combined with spectrogram as crying detectors can achieve the relatively robust performance of 12% over that with MFCC in high-noise environments, which is recommended as a baseline for further work.