Speech as a Biomarker for COVID-19 Detection Using Machine Learning

Abstract

The use of speech as a biomedical signal for diagnosing COVID-19 is investigated using statistical analysis of speech spectral features and classification algorithms based on machine learning. It is established that spectral features of speech, obtained by computing the short-time Fourier Transform (STFT), get altered in a statistical sense as a result of physiological changes. These spectral features are then used as input features to machine learning-based classification algorithms to classify them as coming from a COVID-19 positive individual or not. Speech samples from healthy as well as “asymptomatic” COVID-19 positive individuals have been used in this study. It is shown that the RMS error of statistical distribution fitting is higher in the case of speech samples of COVID-19 positive speech samples as compared to the speech samples of healthy individuals. Five state-of-the-art machine learning classification algorithms have also been analyzed, and the performance evaluation metrics of these algorithms are also presented. The tuning of machine learning model parameters is done so as to minimize the misclassification of COVID-19 positive individuals as being COVID-19 negative since the cost associated with this misclassification is higher than the opposite misclassification. The best performance in terms of the “recall” metric is observed for the Decision Forest algorithm which gives a recall value of 0.7892.

XGBoost-Based Feature Learning Method for Mining COVID-19 Novel Diagnostic Markers

“…Machine learning is applied extensively in biomedical applications, as well as COVID-19 diagnosis ( 8 ). Extreme Gradient Boosting (XGBoost) is a GBDT-based algorithm.…”

Predicting oxygen requirements in patients with coronavirus disease 2019 using an artificial intelligence-clinician model based on local non-image data

“…During the ongoing coronavirus disease 2019 (COVID-19) pandemic, researchers have shown the effectiveness of ML in various fields ( 32 – 34 ); however, potential applications of ML for disease prevention are unclear in real-world medical settings. Furthermore, translational bioinformatics in COVID-19 research has suggested the effectiveness of ML models with customized designs ( 35 ).…”

Omicron detection with large language models and YouTube audio data

“…6 Omicron, however, more commonly affects the upper airway, sinuses, and hypopharynx than prior variants, often resulting in voice changes without a cough. 7 This represents an opportunity for more specific targeting by AI methods if robust datasets were available. Worldwide, there are over 3.6 billion users of various social media platforms, and that number is expected to be above 4.4 billion by 2025.…”

“…Prior AI methods have been unable to successfully detect pre-Omicron variants from unscripted or scripted human voice alone, or have been otherwise unsuitable for deployment (e.g., limited training data, poor generalization) 5 . Omicron variants, however, are typically milder and affect the upper airway more commonly than prior variants, often resulting in voice changes without a cough or other respiratory symptoms 6 . This presents an opportunity for targeting with AI methods, if robust datasets were available.…”

“…[14][15][16][17][18][19] The spectral features of speech in asymptomatic patients with and without COVID-19 yielded a true positive rate of 70%, though the likelihood of generalization was low due to scripted collection and small sample size. 20 In a dynamic pandemic such as COVID-19, crowdsourced datasets allow for continuous sample collection over time. "Coswara" is a database containing respiratory sound samples, including cough, breath, and voices reading standardized scripts.…”