Prior to the pandemic, U.S. schools were experiencing an upward trend in closures due to cold and flu outbreaks. The only tools at their disposal were a thermometer and a school nurse.

But too often, by the time a child visits the nurse’s office, their germs have already spread. Educators needed an early warning system that alerted them in real time that symptoms of an outbreak were on the rise. Nothing like that existed. So, we created a new early warning tool for the classroom.

And then, the COVID-19 pandemic struck. Suddenly, this tool took on new meaning and urgency.

Clorox has kept classrooms clean for generations, but our products are typically used by educators to disinfect surfaces after an illness or mess has occurred.

We set out to invent a tool that employed sound to help identify that something may be spreading on and beyond the surfaces of the classroom. This technology will be the first to detect and analyze the sounds of coughs and sneezes – often the first audible indicators of illness.

While the beta technology was nearing completion, the world was confronted with the COVID-19 pandemic. In addition to building a multi-symptom identification tool, we faced the daunting obstacle of monitoring a room full of students while maintaining their privacy. We needed to show the world a technology that anticipated and met the unique and unprecedented needs of our times.

Inspired by the advancements in voice-driven AI (“Okay Google, play The Weeknd!”), we invented a classroom tool that employs sound -- not to play music, adjust the lighting or tell you when the Mayflower arrived -- but to identify elevated rates of coughs and sneezes.

Our technology provides an early alert system to give educators actionable data never before available indicating that surface germs may be on the rise. The innovative tool uses the power of machine learning and an audio library of tens of thousands of coughs and sneezes to passively analyze a room full of students, and like a smoke detector, alert educators that additional disinfection and other measures may be needed.

Through a custom dashboard, educators receive the information in near real-time that they need to pursue a more active and targeted approach to surface disinfection and classroom protection.

Audio Analysis Tool

Fall 2019

• Developed beta software and learning algorithm to analyze and distinguish between coughs and sneezes from other classroom noises.

• Built the software for a web-based administrative dashboard notification system to view data points.

• Designed and fabricated a prototype tool for testing.

• Initiated field-test in Nova Scotia, Canada and gained data of over 1,300 sound signatures to identify an initial cough and sneeze library of sounds.

• Sound recognition algorithm successfully identified coughs and/or sneezes at a rate of over 95% accuracy.

• Partnered with epidemiologist Dr. Saskia Popescu to detect indicators of symptoms of an outbreak in classroom settings.

• Sent initial tool to Clorox offices for beta testing, but paused due to COVID-19 pandemic, which sent Clorox employees home.

Thermal Capabilities Are Introduced

May 2020

• Began to re-develop prototype to include thermal detection technology and incorporate cough and sneeze listening, all linked to the dashboard notification system.

June-August 2020

• Developed technology that can pinpoint students' temperatures in the classroom through thermal analysis with 95% accuracy rates (in development environment) in identifying a correct temperature assessment.

• Initiated ongoing field-testing in Nova Scotia, Canada helping the device to learn to identify body temperature radiating from the inner eye for the most accurate reading; advanced continued improvement on cough and sneeze learning algorithm.

• Refined research proposal with epidemiologist Dr. Saskia Popescu to include a threshold for fevers in the classroom setting.

August 10, 2020

• Began 3-D printing of initial prototype to house both the cough and sneeze software and the addition of a stand to help measure temperature at the front of the classroom; this ensures the tool is at an optimal height to capture body temperature at the inner eye.

• Developed dashboard to process algorithm results that could report back to teachers and schools.

• Drafted copy that would accompany alerts from the dashboard to teachers and school administrators when an increase in coughs and sneezes is detected, which may disperse germs on classroom surfaces.

August 21, 2020

• Filmed an informational video that explains how the tool works.

September 18, 2020

• Conducted a field test, proving beta technology identified coughs and sneezes with 95% accuracy.

September 22, 2020

• Built microsite to serve as a place to share an informational video and offer the opportunity to sign up to learn more information.

• Created first tangible model of TASC (Tool for Audio Sneezes and Coughs) via 3-D printing to use for future launch announcement.

September 28, 2020

• Conducted another field test, including a demonstration using recorded audio noises and a test subject in a heated blanket to simulate body temperature.

• Field tested dashboard to show transfer of data from the tool in real time to the alert system that notifies educators.

November 2020

• Continued to refine proprietary cough, sneeze and thermal algorithms to reach superior performance in testing.

December 2020 – ongoing

• Researching necessary regulatory testing/compliance needs to determine feasibility to bring to market.

•The tool is placed near regular student foot traffic. The microphone picks up sounds within a 15-ft. radius that are digitized into the CPU via cellular modem.

•The CPU converts the digitized sound signal into multiple time-sliced spectrograms that are convolved over sliding time windows against a trained spectrogram machine learning model of what a cough looks like.

•The model uses two datasets consisting of 50,000 sounds, one of cough and sneeze spectrograms and one of general classroom spectograms like laughter, chatter and moving chairs. The team gathered audio from North American households to account for different languages, ages, ethnicities and voice frequencies.

•Time slices are processed and analyzed by the machine learning model. An algorithm is used to raise a single flag for each cough. High scores are flagged as a “hit” and that cough or sneeze is logged to the database.

•An AI feature senses thermal pixels at the tear duct. If a reading reaches 100.4°F or above, it registers as an alert. No visuals are captured to ensure anonymity.

•If the tool detects symptoms that surpass a predetermined threshold developed in consultation with an epidemiologist, administrators receive notifications alerting them to evaluate the classroom.

As a brand that has spent a century helping classrooms remain clean, safe spaces for learning, Clorox started with a simple question, “Could we be doing more?” And “What might that look like if we harnessed the power of technology and machine learning?”

An initial field test revealed that our innovative beta technology can identify coughs and sneezes with over 95% accuracy – which earned it an enthusiastic response.

Over 1,000 people signed up on our microsite in one week, eager to learn more information as development continues.

With that, our “what if” beta tool went from theoretical test subject to a catalyst for Clorox to explore how data-enhanced technologies to help detect symptoms of disease. We anticipate thousands of schools nationwide will be interested.