Emotion Recognition of Infant Cries using Multi-Scale CNN-BLSTM

Infant cry is a speech signal that carries rich emotional information, which can not only express the physiological needs of infants but also reflect their emotional states. Currently, there is no publicly available dataset for research on infant cry emotion recognition. To address this issue, we collected and selected 5242 infant cry samples from the internet as a dataset for infant cry emotion recognition. In addition, we proposed a Multiscale Convolutional Neural and Bi-directional Long Short-Term Memory Networks (Multiscale CNN-BLSTM) based method for infant cry emotion recognition. Experimental results show that our proposed model achieved an emotion recognition rate of 83%, which is nearly 11% higher in accuracy compared to convolutional neural networks without multiscale technology.