Krisp app is based on Deep Neural Networks. We apply various data augmentations to cover microphone diversity, acoustic conditions, signal-to-noise ratios, bandwidths, and many other factors.
We have collected and listened to datasets of 20K distinct noises and 10K clean voices of different ages, gender, and ethnicity.
We had performed thousands of hours of listening projects to identify, label and clean the audio datasets. During the ML trainings, our models saw around 170 years of diverse audio speech.
Learn more about Krisp technology in our Engineering Blog.