COVID-19 Detection Using Coughing Sounds with Mel-Frequency Cepstral Coefficients and Long Short-Term Memory

Lim, Jia Chong (2023) COVID-19 Detection Using Coughing Sounds with Mel-Frequency Cepstral Coefficients and Long Short-Term Memory. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Full Text -Lim Jia Chong.pdf Restricted to Registered users only Download (3MB)

Abstract

As there are a lot of limitations on current existing approach in screening of COVID-19 infection, an efficient approach must be introduced to the healthcare application as soon as possible in order to inhibit the spreading chain of COVID-19 around the world. Human can listen to audio file, but could not interpret the audio signal precisely. However, computers with deep learning algorithm could do so while handling huge amount of data. Therefore, the main focus of this research project is to develop a deep learning model in detecting COVID-19 infection through the analysis of coughing sound, Long Short-Term Memory (LSTM) is used as the deep learning neural network in this research project. It is an improved version of recurrent neural network (RNN) and it is specialized in processing time-series data which is also known as audio signals. As a result, the aim of this research project is to build a LSTM model with Mel-Frequency Cepstral Coefficients (MFCCs) feature as a diagnostic tool for COVID-19 infection. In order to achieve this, Coswara database is utilised as the source of coughing dataset, the coughing dataset is then go through the pre-processing process and hence employed for the model learning and training. Lastly, the trained model has achieved an accuracy of about 58% and its feasibility was evaluated with an unseen test dataset based on the classification report metrics.