The Brief Explanation
Your audio input is converted into a visualization of a Fast Fourier transform (FFT) that a Convolutional Neural Network (CNN) can then use to identify key components of the cough. Once the CNN makes a prediction on a strength (classified between 0 and 9), a multivariable linear regression algorithm predicts the image of the cough.
This prediction is based on over a hundred cough images processed into a pixel array with thousands of individual comparisons. In total, the predicted image is created by a machine learning model that has gone through over a million total comparisons.
Once the prediction is complete, each individual pixel's HSB color value is used to identify key components of the cough. Values below a certain threshold are cut out to distinguish the cough from the rest of the image, and border pixels were used to calculate the cough's breadth and length. The higher the pixel value, the more dense and larger the particles in that pixel are, which is used to calculate the regions within a cough that are more likely to transmit the virus and/or bacteria.
Convolutional Neural Networks are a class of deep learning neural networks that analyzes images.
Findings were correlated with data from the Lancet Medical Journal, and we found remarkable similarities.
We are two 17 year old high school students who set out to democratize cognitive science for non-research applications.
Check out the numerous educational graphs, guides, videos, and other resources we've made to help you understand our technology.
This site helps individuals understand their personal risk of transmitting COVID-19, and data collected can be repurposed for science research and policy-making.
Learn More >