Google has unveiled Lyra, a new technology that enables Duo, and soon other applications, to deliver natural-sounding voice chat with just 3 kilobytes per second of network bandwidth.
Update: Google has now made Lyra open source for other applications to use.
With many of us still unable to visit loved ones, video calling apps have played a key role in keeping us all connected to each other, with Google Duo and Meet hosting over 1 trillion minutes of video in the last year alone. However, this has also put great pressure on the internet infrastructure around the world, and most video calling methods exclude anyone with a low quality internet connection.
To help with this, Google has developed a new audio codec called Lyra that is specifically optimized to deliver natural-sounding, understandable, and recognizable human speech in the smallest possible space. As explained on the Google AI Blog, this was done with the help of a machine learning model, which was trained in “thousands of hours of audio with speakers in over 70 languages” to ensure that Lyra could be used by as many people as possible. More importantly, Lyra is efficient enough to be able to run on anything from a high-end cloud server to a mid-range smartphone with just 90ms latency.
As you can see, or rather hear, in the video above, Lyra offers audio that, while notably lower quality than a normally encoded recording, is clearly recognizable as the speaker’s voice while still managing to use an incredibly large amount. small data. There are a few more speech samples in Lyra compared to other low-bandwidth audio codecs, on Google’s AI Blog.
Lyra will soon get its first real-world use, as it is now being rolled out to Google Duo for Android, where the codec will be used for calls made over low-speed connections, and Google will pay special attention to dial-up connections and connections. Rural areas in India and Brazil only offer a 2G network connection. From there, Google plans to release Lyra as open source, allowing other companies to offer low-bandwidth Lyra audio in their own applications.
Update 4/6: Just over a month after being first announced, Google has delivered on its promise and made Lyra completely open source. For now, this first beta version of the Lyra codec is only optimized for use by Android developers on Linux machines, but it should be enough to give developers something to start working with and then bring them to all of their intended platforms.
Today we released Lyra as a beta version because we wanted to enable developers and get feedback as soon as possible. As a result, we expect the API and bitstream to change as they develop. All code to run Lyra is open source under the Apache license, except for a math kernel, for which a shared library is provided until we can implement a completely open solution on more platforms. We hope to see what people do with Lyra now that it is open source. Check out the code and demo on GitHub, let us know what you think and how you plan to use it.
More about Google Duo:
FTC: We use income generating automobile affiliate links. Plus.
Check out 9to5Google on YouTube for more news: