The Krisp app is built upon Deep Neural Networks. We employ a range of data augmentations to encompass microphone diversity, acoustic conditions, signal-to-noise ratios, bandwidths, and various other factors.
We have gathered and reviewed datasets comprising 20,000 unique noises and 10,000 clear voices from individuals of diverse ages, genders, and ethnic backgrounds.
We have undertaken numerous hours of listening projects to identify, categorize, and refine audio datasets. Throughout the machine learning training process, our models were exposed to approximately 170 years of diverse audio speech.
Learn more about Krisp technology in our Engineering Blog.