Google LLC said today it has open-sourced a codec called Lyra that relies on machine learning to enable high-quality voice calls in low bandwidth situations.

Lyra, which is now hosted on GitHub, can compress audio down to as little as 3 kilobits per second while still ensuring a sound quality that compares well with other codecs that require much greater bandwidth.

Writing about Lyra for the first time in February, Google said that although mobile connectivity has spread rapidly across the world, not everyone has access to fast and reliable internet connections. Moreover, the emergence of remote work and telecommuting during the COVID-19 pandemic has meant that data limits are being stretched even in areas with more reliable connections, Google said. In the U.S., for example, 90 of the top 200 cities saw a significant decline in internet speeds at the start of the panic, according to an analysis from BroadbandNow.

Google said today it is making Lyra open-source to help make a difference in these kinds of situations.

The Lyra codec has been optimized to squeeze recognizable and natural-sounding human speech into as small a space as possible. Google said Lyra was built using a machine learning model trained on thousands of hours of audio of people speaking, and in more than 70 languages, to ensure it can be used by as many people as possible. Just as important, the codec can run on low-powered devices, including a smartphone that only has 90 milliseconds of latency, Google said.

Google explains on its open source blog that Lyra’s architecture has two basic components – an encoder and a decoder. The encoder is used to capture the distinctive attributes, or features, of someone’s speech when they are speaking into the phone. Lyra grabs these features in 40-millisecond chunks, compresses them, then sends them over the network. At the other end, the decoder converts those features back into an audio waveform that can be received and understood by the person at the other end of the call.

“Whereas these traditional codecs are based on digital signal processing (DSP) techniques, the key advantage for Lyra comes from the ability of the generative model to reconstruct a high-quality voice signal,” Google Chrome engineers Andrew Storus and Michael Chinen wrote in a blog post.

Here’s an example of the call quality when using Lyra. It’s noticeably lower-quality than what many people are used to with a normally encoded recording, but it is still distinctly recognizable nevertheless.

Google Duo’s new audio technology Lyra gives people high-quality and reliable audio, even on a 2G network. Hear more from #Android product manager Jamieson Brettle. pic.twitter.com/asAkV3Gi8s

— Android (@Android) March 2, 2021

“Being the provider of the number one smartphone OS with Android, Google is doing a very good job helping the less bandwidth-rich parts of the world enjoy a better user experience,” said analyst Holger Mueller of Constellation Research Inc. “Voice is a much more natural way for humans to communicate than any screen and so Lyra is a key step to bring that in to low bandwidth environments. Open-sourcing Lyra is a good move as it creates the transparency and trust for third party developers to use the service.”

Google has already implemented Lyra in its free Duo video calling app, and it said it’s open-sourcing the code because it believes it might be suitable for more applications. Google believes there are a number of applications to which Lyra might be uniquely suited, from archiving large amounts of speech and saving battery to alleviating network congestion in emergency situations.

“We are excited to see the creativity the open source community is known for applied to Lyra in order to come up with even more unique and impactful applications,” Storus and Chinen said. “We [want] to enable developers and get feedback as soon as possible.”

Image: Google