Detection of COVID-19 by Cough Sound: A Method Based on DSC + BiLSTM

In recent years, the outbreak of COVID-19 has brought a new round of challenges to global health care, and daily large-scale testing has also increased the consumption of medical resources. However, studies have shown that the cough sounds of patients with COVID-19 are significantly different from other Characteristics of respiratory infectious diseases. Therefore, this paper considers the use of the patient’s cough as a detection sample to give the preliminary screening results. The research was conducted on the COUGHVID dataset. The experiment is divided into two stages: (1) Preprocessing stage: use Pitch Shift and Time Stretch to perform data enhancement on audio data, and use spec Augment to perform data enhancement on mel spectrogram. (2) Model construction stage: use two layers of DSC and one layer of BILSTM to splicing to obtain a classification model. Finally, the method is compared with the baseline method using only two layers of LSTM. The results show that accuracy has increased by 1.9%, F1 has increased by 1.9%, and AUC has increased by 1.6%.