There’s a slim chance anyone has managed to avoid video or voice calling services that rely on internet connections by this stage, thanks in no small part to the global pandemic.
There’s an even slimmer chance that these calls have been entirely free from dropouts, glitches, latency and other disruptive artefacts, but the latest advancement from Google’s AI division should help out on this front.
Lyra’s improvements are the result of Google AI’s combination of existing codec technology and “advances in machine learning with models trained on thousands of hours of data.” This data includes speakers in over 70 languages, freely available in open source libraries, to ensure its universal applicability.
The blog itself goes into much more detail about how this machine learning enabled the new technology, and also uses audio clips to demonstrate the difference between the original audio, Lyra, and other codecs at different bitrates.
The team plans to continue optimizing Lyra and improving its quality and reliability in hopes that it will be used outside of the Google Duo app. This will likely mean that other VOIP and video conferencing apps and services will benefit from the codec in the future.
In addition, the message states that the team “is also beginning to explore how these technologies can lead to a low-bitrate general-purpose audio codec,” which could mean that non-voice use cases such as music and other audio could see the same improvements.
Google Duo call quality enhances for users with slow internet thanks to AI