Detection of Covid-19 Through the Analysis of Vocal Fold Oscillations

Mahmoud Al Ismail, Soham Deshmukh, Rita Singh

TL;DR

Researchers have developed a method to detect COVID-19 by analyzing the biomechanical vibrations of the vocal folds. By using a dynamical system model to reconstruct vocal fold motion from simple voice recordings, the team found that COVID-19 significantly disrupts the symmetry and synchronization of speech production, allowing for detection with high accuracy (up to 0.91 AUC) using basic machine learning classifiers.

What Problem Does This Paper Address?

While many researchers have attempted to detect COVID-19 from voice and cough recordings, most rely on "black-box" deep learning models. These models often struggle with small datasets and are prone to overfitting—learning speaker-specific quirks rather than actual disease markers.

This paper takes a physics-first approach. COVID-19 is primarily a respiratory illness that impairs lung function and airway musculature. Since the "source" of our voice (phonation) relies on an intricate balance of aerodynamic forces across the vocal folds, the authors hypothesized that the disease would leave a measurable "signature" in how those folds vibrate. The challenge lies in the fact that we cannot easily film someone’s vocal folds in a clinical setting at scale; therefore, we need a way to mathematically "see" those vibrations through recorded audio.

Methods

The authors utilize a 1-mass asymmetric body-cover model, a mathematical framework that describes the horizontal displacements of the left and right vocal folds using coupled non-linear differential equations.

To bridge the gap between raw audio and these physical equations, they employ the ADLES (Acoustic-to-Dynamic Lossless Estimation of Speech) algorithm. This approach jointly solves two problems:

  1. The Forward Problem: Predicting vocal fold motion based on physical parameters (like mass, spring tension, and asymmetry).
  2. The Inverse Problem: Estimating those physical parameters by comparing the model’s predicted glottal flow to the actual flow extracted from a voice recording via inverse filtering.

By minimizing the "residual energy" (the difference between the model and the real-world signal), they can extract features like the coupling coefficient ($\alpha$), the asymmetry coefficient ($\Delta$), and the residual error ($R$) itself.

Datasets

The study used a clinically curated dataset from Merlin Inc. (Chile).

  • Total Pool: 512 individuals tested for COVID-19.
  • Selected Subset: 19 individuals (10 female, 9 male) who were recorded within 7 days of their medical test.
  • Breakdown: 9 COVID-positive and 10 COVID-negative subjects.
  • Audio: Recordings of isolated extended vowels (/a/, /i/, and /u/) sampled at 8 kHz.

Key Insights and Ideas

  • Phase Space Disruption: In healthy individuals, vocal fold trajectories in phase space (displacement vs. velocity) look like "limit cycles" or "slim toroids"—indicating highly synchronized, symmetrical motion. In COVID-19 patients, these trajectories become complex and "messy," signaling a loss of synchronization and a reduced range of motion.
  • The Power of the Residual: The most discriminative feature wasn't just the physical parameters, but the residual $R$. This suggests that the standard mathematical model for healthy speech simply cannot account for the "anomalous" vibrations caused by the infection. The model's "failure" to fit the data becomes the biomarker.
  • Vowel Sensitivity: The vowel /i/ (a high front vowel) was significantly more effective for detection than /a/ or /u/. The authors suggest this is because producing /i/ requires reaching specific high-frequency energy peaks that are particularly compromised by respiratory impairment.

Key Results

The researchers tested several classifiers, including Logistic Regression (LR), Support Vector Machines (SVM), and Random Forests (RF).

  • General Performance: Most classifiers achieved an area under the curve (ROC-AUC) of approximately 0.80 across all vowels.
  • Top Performer: Logistic Regression using only the vowel /i/ achieved a remarkable AUC of 0.912.
  • Statistical Significance: All results were found to be highly significant, with p-values better than $1e^{-5}$, suggesting that these biomechanical features are robust indicators of the disease.

Limitations and Future Work

The authors are transparent about two major caveats:

  1. Specificity: The study does not yet prove that these vocal anomalies are exclusive to COVID-19. Other respiratory conditions (like the flu or pneumonia) might produce similar disruptions in vocal fold entrainment.
  2. Symptomatic Bias: The method is likely only effective for symptomatic patients whose respiratory functions are already impaired. It may not serve as a tool for detecting asymptomatic carriers.

Future work involves using more sophisticated dynamical models to better capture the "pathological" oscillations and expanding the dataset to test the system against other respiratory illnesses to ensure true diagnostic specificity.

Summary updated

Abstract

Phonation, or the vibration of the vocal folds, is the primary source of vocalization in the production of voiced sounds by humans. It is a complex bio-mechanical process that is highly sensitive to changes in the speaker's respiratory parameters. Since most symptomatic cases of COVID-19 present with moderate to severe impairment of respiratory functions, we hypothesize that signatures of COVID-19 may be observable by examining the vibrations of the vocal folds. Our goal is to validate this hypothesis, and to quantitatively characterize the changes observed to enable the detection of COVID-19 from voice. For this, we use a dynamical system model for the oscillation of the vocal folds, and solve it using our recently developed ADLES algorithm to yield vocal fold oscillation patterns directly from recorded speech. Experimental results on a clinically curated dataset of COVID-19 positive and negative subjects reveal characteristic patterns of vocal fold oscillations that are correlated with COVID-19. We show that these are prominent and discriminative enough that even simple classifiers such as logistic regression yields high detection accuracies using just the recordings of isolated extended vowels.
BibTeX
@inproceedings{ismail2021detection,
  author = {Mahmoud Al Ismail and Soham Deshmukh and Rita Singh},
  title = {{Detection of Covid-19 Through the Analysis of Vocal Fold Oscillations}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2021},
  url = {https://openreview.net/forum?id=1e67m3y0pu},
}

Loading related content...

Figure