COVID-19 Detection Algorithm Using Voice Quality Features and VLAD Coding

In 2020, the World Health Organization declared the COVID-19 outbreak a pandemic. In order to promote the rapid and reliable detection of COVID-19, this research introduced voice signal processing technology to find the voice signal characteristics of COVID-19 infection, and automatically judges whether it is infected with COVID-19 using cough fragments and speech fragments. On the basis of the relevant data set and baseline provided by INTERSPEECH 2021ComParE, firstly, the audio segmentation technology was used to augment the data set. And secondly, voice quality features were added to the feature set, which improved baseline results and proved that the voice quality features are effective on the task of automatic speech detection for COVID-19. At the same time, Vector of Locally Aggregated Descriptors is introduced to encode low-level features. When the dictionary size is small, the classification performance of the system is effectively improved. Finally, the classification results obtained by multiple algorithms are fused to further improve the final classification effect. The UAR for CCS and CSS sub-challenges are 73.9% and 77.2%, respectively.