Google has announced Lyra, a new technology that will enable Duo and other apps to provide natural-sounding voice chat with as little as 3 kbps of network bandwidth.

On Android, Google Duo uses ‘Lyra' to provide natural-sounding voice calls even on a 2G connection. [Updated]


Lyra is also open source and can be used for other users, according to Google.

With too many of us still unable to visit with loved ones, video calling apps such as Google Duo and Meet have played an important role in keeping us all connected, with Google Duo and Meet alone hosting over 1 trillion minutes of video last year. However, this has placed a significant burden on global internet connectivity, and most video calling solutions completely preclude those with a poor internet connection.

To aid in this, Google has created Lyra, a modern audio codec designed to provide familiar, readable, and natural-sounding human speech in the smallest amount of space available. This was accomplished with the help of a machine learning algorithm that was educated on “thousands of hours of audio with speakers in over 70 languages” to ensure Lyra could be used by as many users as possible, according to the Google AI Blog. More specifically, Lyra is powerful enough to run with just 90ms of latency on anything from a high-end cloud server to a mid-range handset.

Lyra provides audio that is clearly identifiable as the speaker's voice by using very little data, as seen — or more heard — in the video above. While the audio is of slightly poorer quality than a typically encoded file, it is distinctly recognizable as the speaker's voice while using extremely little data. On the Google AI Blog, there are some more voice samples in Lyra relative to other low-bandwidth audio codecs.

Lyra will shortly see its first real-world use, as the codec is now available for use in Google Duo for Android, where it will be used for calls made over low-speed networks, with Google highlighting dialup connections and remote areas in India and Brazil that only have a 2G network service. Following that, Google intends to open source Lyra, encouraging other businesses to use low-bandwidth Lyra audio in their own applications.

Update 4/6: Google has maintained its pledge and made Lyra totally open source, just over a month after it was first revealed. This first beta version of the Lyra codec is only designed for Android developers on Linux machines for the time being, but it should be enough to get developers started and eventually introduce them to all of their planned platforms.

We're launching Lyra as a beta today because we want to get input from users as soon as possible. As a result, we anticipate that the API and bitstream will evolve as the project progresses. But for a math kernel, which has a shared library before we can apply a more open solution across more platforms, all of the code for running Lyra is open-sourced under the Apache license. We're excited to see what people do with Lyra now that it's free to use. Visit GitHub to see the code and tutorial, and let us know what you think and how you want to use it!


Post a Comment

Previous Post Next Post