May 18, 2020 — Audio signals generated
by the human body have often been used by clinicians and clinical researchers as diagnostic or progression
indicators for diseases and disease onset. The human body’s catalogue of sounds includes sighs, breathing,
the heart, digestive system and vibration sounds, as well as voice.
Until recently, such signals were usually collected with manual stethoscopes
during scheduled medical appointments. Research has now started to use digital technology to gather bodily
sounds, and run automatic analysis on the data. For example, digital stethoscopes developed by Littmann
and Thinklab have aided detection of asthma and wheezing.
Researchers have also been piloting the use of the human voice
as a proxy for early diagnosis of a variety of illnesses. Parkinson’s disease correlates to softness of
speech, resulting from lack of coordination over the vocal muscles1, 1b. Coronary artery
disease can be indicated by voice frequency, as hardening of the arteries may affect voice production
2. Invisible injuries like post-traumatic stress disorder3, traumatic brain injury
and psychiatric disorders4, 4b correlate to changes in the tone, pitch, rhythm, rate and volume
of the human voice.
The use of human-generated audio as a biomarker for various illnesses offers
enormous potential for early diagnosis. The technology could offer an inexpensive solution with the
potential to be rolled out on a mass scale, if it could be embedded in commodity devices. This is even
more true if such solutions could monitor individuals throughout their daily lives in an unobtrusive way.
Why have we initiated data collection of audio sounds for COVID-19?
Respiratory sounds are already known to have diagnostic
properties5. Moreover, there is some initial evidence that respiratory sounds, especially
coughing, could be indicative of COVID-19 6.
However there is insufficient data to determine whether this is
generally true, and to date there is no research investigating whether breathing patterns, voice features
and coughs could help the diagnosis of COVID-19. While the single audio features might not serve as robust
diagnostics alone, we want to see whether they could be powerful taken all together, and add information
to the self-declared symptoms of the individual.
We have therefore initiated a data collection which collects
the sounds of voice, breathing and coughing from participants, via our COVID-19 Sounds
What do we hope to achieve?
A large sounds dataset from users across different demographics
and with different medical histories, containing thousands of samples crowdsourced in the “wild”, is
We hope to develop models which use the audio inputs to see
whether COVID-19 could be detected primarily through audio. We believe the most promising avenue of this
is the detection of disease progression. Having multiple samples of individuals at different stages might
allow researchers to build a better picture of COVID-19’s progression in our respiratory
Secondly, this effort will offer the community a dataset for
research into COVID-19 and, as our dataset grows, perhaps it will also be of use in research into other
 R. Vikas and R. Sharma. Early detection of Parkinson's disease through
Voice. 2014 International Conference on Advances in Engineering and Technology (ICAET), Nagapattinam,
2014, pp. 1-5.
[1b] Brabenec L, Mekyska J, Galaz Z, Rektorova I. Speech disorders in Parkinson's disease: early
diagnostics and effects of medication and brain stimulation. J Neural Transm (Vienna). 2017
Mar;124(3):303- 334. doi: 10.1007/s00702-017-1676-0. Epub 2017 Jan 18.
 E. Maor, J. Sara, D. Orbelo, L. Lerman, Y. Levanon, A. Lerman . Voice Signal Characteristics Are
Independently Associated With Coronary Artery Disease. Mayo Clin Proc. 2018 Jul;93(7):840-847. doi:
10.1016/j.mayocp.2017.12.025. Epub 2018 Apr 12.
 D. Banerjee, I. Debrup, A. Kazi, G. Mei, L. Xiao, G. Zhang, R. Xu, S. Ji, J. Li. (2017). A Deep
Transfer Learning Approach for Improved Post-Traumatic Stress Disorder Diagnosis. In Proceedings of Int.
Conference on Data Mining (ICDM’17).
 D. Vergyri, B. Knoth, E. Shriberg, V. Mitra, M. McLaren, L. Ferrer, P. Garcia and C. Marmar.
Speech-based assessment of PTSD in a military population using diverse feature classes. Interspeech, 3729-
[4b] M. Faurholt-Jepsen, J. Busk, M. Frost, M Vinberg, E. Christensen, O. Winther, J. Bardram, L. Kessing.
Voice analysis as an objective state marker in bipolar disorder. Translational Psychiatry. volume 6, page
 Pramono RXA, Bowyer S, Rodriguez-Villegas E (2017) Automatic adventitious respiratory sound analysis:
A systematic review. PLOS ONE 12(5): e0177926. https://doi.org/10.1371/journal.pone.0177926
 A. Imran, I. Posokhova, H. Qureshi, U. Masood, S. Riaz, K. Ali, C. John, M. Nabeel, I. Hussain.
AI4COVID-19: AI Enabled Preliminary Diagnosis for COVID-19 from Cough Samples via an App.
 B. Schuller, D. Schuller, K. Qian, J. Liu, H. Zheng, X. Li. COVID-19 and Computer Audition: An
Overview on What Speech & Sound Analysis Could Contribute in the SARS-CoV-2 Corona Crisis.