The Mozilla Research RRNoise project shows how to apply deep learning to noise suppression. It combines classic signal processing with deep learning, but it’s small and fast. No expensive GPUs required — it runs easily on a Raspberry Pi. The result is easier to tune and sounds better than traditional noise suppression systems (been there!).

RNNoise will help improve the quality of WebRTC calls, especially for multiple speakers in noisy rooms. It is also small enough and fast enough to be executed directly in JavaScript, making it possible for Web developers to embed it directly in Web pages when recording audio.

Screenshot of the player that evaluates the effect of RNNoise. Play the demo.

You can improve RNNoise by donating your noise to science. We’re interested in noise from any environment where you might communicate using voice. That can be your office, your car, on the street, or anywhere you might use your phone or computer. The more realistic noise we have, the better the models we can build and the better the output.

Read in depth about the RNNoise project.