Opus 1.5 achieves continuous, audible speech at 90% packet loss

The free and open Opus audio codec released version 1.5. The release notes contain a surprising amount of machine learning features and descriptions. The most surprising one, being demonstrated with audio: audible speech at 90 % packet loss.

90 % of data gets lost, but with the data and machine learning model Opus can fill the 10 % data into continuous, audible speech. It does sound jittery, but is a truly impressive feat.

Section WebRTC Integration → Samples has an example where you can test out the 90 % packet loss audio.