IISC developed COSWARA helps diagnose COVID19 through voice samples

Indian Institute of Science has never ceased to surprise us with its studies and inventions even during the pandemic time. Dr. Sriram Ganapathy, Assistant Professor in the electrical engineering department of IISC has recently developed a technology, COSWARA, which will detect COVID19 infection using voice recognition. Speaking exclusively to Prashobh Devanahalli, ETV Bharat's Reporter in Bengaluru, Dr. Sriram explained the development and use of the technology.

Bengaluru: Indian Institute of Science has never ceased to surprise us with its studies and inventions even during the pandemic time. Dr. Sriram Ganapathy, Assistant Professor in the electrical engineering department of IISC has recently developed a technology, COSWARA, which will detect COVID19 infection using voice recognition. Speaking exclusively to Prashobh Devanahalli, ETV Bharat's Reporter in Bengaluru, Dr. Sriram explained the development and use of the technology.

Here are some excerpts from the interview.

Prashobh: Dr. Sriram could you briefly tell us what is COSWARA and how does it help in identifying COVID19 related infections?

Dr. Sriram: Namaste everyone, thank you for the opportunity to talk to you regarding COSWARA.

The name COSWARA is a combination of COVID and SWARA which mean sound. This sound-based tool helps in the diagnosis of COVID19. Participants need to use smartphones or web-connected devices to record some sound of their breathing, coughing and speech. They also need to indicate their health status. Subsequently, this data is uploaded to a remote server and we analyse the data using some algorithms and try to come up with some hypotheses. We try to distinguish between the sound of a healthy person and COVID infected individual. Thanks to the efforts of several of my fellow colleagues, medical professionals and authorities of hospitals associated with us during the research.

The efforts to develop this took started about a year ago. We crowdsourced data from individuals from remote locations. We gathered the data and analysed it. The idea was to cater the results of the analysis to the users through their cell phones or a website or through an App. We put together data and algorithms to form hypotheses.

Prashobh: How will a computer or software evaluate the sound samples to arrive at a diagnostic outcome?

Dr. Sriram: Let us try to understand what the COVID19 infection does to the human body. When a person is exposed to the virus, it incubates in the affected individual's respiratory tract. The virus gets to the lungs of the patient through the oral and nasal cavities. It will grow in the lungs of the patient. As you may very well know, diagnostic methodologies like CT Scan and X-ray show the various characteristics of COVID19 infection. The infection will affect the sounds generated by the respiratory system itself. Speech, breathing sound and coughing sounds are the results of the coordinated activity of various organs in the respiratory system.

So, when there is an infection in the respiratory organs, the characteristics of the sound generated by these parts alter. For example, the sound of the cough in COVID19 participants has a distinct vibratory nature. With enough samples with sound characteristics from COVID19 infected individuals and sounds from healthy people, we can devise a computer algorithm to extract the right features which will indicate infection. The data gathered from a number of COVID positive and negative samples will help us to map new samples to a probability score, which can say that a participant taking the test has a certain percentage of chance of being infected.

Prashobh: So far how many voice samples have been collected for the development of COSWARA tool?

Dr. Sriram: We have collected about 2,000 samples. Thanks to the volunteers who contributed the data. I request all to kindly go to the website coswara.iisc.ac.in to contribute data. It will take about five to seven minutes. The website will collect details about COVID19 and current health status of the users. It will also collect voice and cough samples. There is no personal data collected. We use it to build our tools further based on the data collected. The more data we collect the tool which is based on machine learning and Artificial intelligence (AI) will improve in terms of accuracy. We have observed that over the course of data collection, the analyses by our tools have improved. We are making great progress.

Prashobh: How accurate is COSWARA?

Dr. Sriram: We have put out the tool and research behind it for peer review. We have reported nearly 93 percent accuracy. The current accuracy percentage can be dissected into two forms. One is how accurately the algorithm was able to detect COVID19 infections among the participants. The second measure is how accurately the algorithm ruled out COVID19 possibility in participants. These are two measures that are used to evaluate the performance of the diagnostic tools. Indian Council of Medical Research has drawn up guidelines to measure the performances of tools such as COSWARA. The guidelines have set a benchmark for each of the two measures for the regulatory model to approve it. So the current benchmark, as of May 2021, the point-of-care tests—the remote tools such as COSWARA—says that they need to be 95 percent accurate when it comes to true negatives. The guidelines also mandate more than 50 percent sensitivity of the point-of-care-test models.

Our tool has met the benchmark at 95 percent and 69 percent when it comes to the sensitivity of the test model. We are almost above 19 percent above the benchmark to qualify as a point-of-care tool. We will do more validations and if the numbers and the results hold up, we would like to bring it to the ICMR's notice with a formal application.

Prashobh: Is there any plans to collaborate with the industry for wider use?

Dr. Sriram: We have been interacting with stakeholders from the industry over the course of the development of this tool. If the tool meets the criteria set by the government regulatory bodies, then we can rely on the industry for help to set up a mobile application that helps in remote testing. We can also come up with more centralised ways in which more people can jointly do the testing. These kinds of massive scaling up of technology to population-level need industry participation.

Currently, about 2,000 people have contributed data through the tool we have developed. However, we don't have the technical capacity to host an application that could handle a situation in which hundreds of people simultaneously try to access it. For an increased technical capacity, we would like to take the help of the industry for capacity building.