Detection of COVID-19 from speech signal using bio-inspired based cepstral features

Highlights：

• Identification of the best possible sound patterns during coughing, breathing, and voiced sounds to efficiently detect COVID-19.

• Application of the speech enhancement schemes for the improvement in the classification performance.

• Use of the Adaptive Synthetic Sampling Approach for Imbalanced Learning to remove the class imbalance in the database and to evaluate properly the various performance measures of the proposed classifier.

• Using the same classifier, comparison of the detection performance of the proposed cepstral features as inputs with different existing cepstral features employing two separate standard databases.

• Demonstration of overall 5% enhancement in detection performance compared to that of other four existing features based method.

摘要

•Extraction and analysis of the cepstral features used in speech recognition. The optimization of the conversion scale in the frequency domain, and the frequency range of filter bank using the bio-inspired technique to facilitate COVID-19 detection.•Identification of the best possible sound patterns during coughing, breathing, and voiced sounds to efficiently detect COVID-19.•Application of the speech enhancement schemes for the improvement in the classification performance.•Use of the Adaptive Synthetic Sampling Approach for Imbalanced Learning to remove the class imbalance in the database and to evaluate properly the various performance measures of the proposed classifier.•Using the same classifier, comparison of the detection performance of the proposed cepstral features as inputs with different existing cepstral features employing two separate standard databases.•Demonstration of overall 5% enhancement in detection performance compared to that of other four existing features based method.