Cough-based COVID-19 Detection with Multi-band Long-Short Term Memory and Convolutional Neural Networks

ABSTRACT

Cough-based COVID-19 detection has shown competitive results through artificial intelligence. In this paper, we proposed a cough-based COVID-19 detection method that made full use of frequency information at different stage. In the feature extraction stage, we proposed band weighted Mel-Frequency Cepstral Coefficients to emphasize features at different frequency bands;in the classification stage, we proposed a multi-band Long-Short Term Memory Convolutional Neural Network with attention mechanism to capture detailed features in the frequency domain. We also combined SpecAugment and Mixup to improve the generalization ability of our proposed model. We evaluated the performance of our proposed model on the dataset of DiCOVA challenge 2021 and our collected dataset. Experimental results showed that the AUC of our model outperformed the first place of DiCOVA challenge 2021 by 5.11% absolutely on average. © 2021 ACM.