Go to website
If you want to dive deep behind the magic of Krisp
How did you make the technology
Krisp app is based on Deep Neural Networks. We have collected and listened to datasets of 20K distinct noises and 10K clean voices of different age, gender and ethnicity. All the audio together compile up to 2.5K hours. The datasets were fed to the neural network which was later trained to remove background noise and leave only clean voice. Refer to this article published on NVIDIA blog by our team:
What data are you sending to your cloud
Your privacy is important to us. All the audio processing during the call is done locally on the device - it is not sent to or stored on any clouds or servers. We only store analytical events to monitor and improve your experience.
How is your app working
Krisp adds an additional layer between the physical microphone/speaker and conferencing app of your choice. To remove the noise coming from other participants to you (Krisp Speaker), the audio is received from the conferencing app, processed on your device to remove noise and then transmitted to your physical speakers. To remove the noise going from you to other participants (Krisp Microphone), the audio is received from your physical microphone, processed on your device to remove noise and th