As people deal with the coronavirus pandemic around the world, they are relying more and more on the online modes of communication to remain connected with friends and family. In such a situation, it is imperative that the communication tools they use are able to offer good experience. One such tool is Duo that is used by people to video chat with their contacts. Google on Wednesday revealed that it has been working for some time to improve the audio quality of Duo calls using machine learning. The fruits of Google’s efforts are already available to Pixel 4 users, but the company says the same technology will reach other phones soon.

According to a blog post on Google’s AI blog, Google has developed something called WaveNetEQ that is packet loss concealment system based on machine learning. Essentially, when you make a call online, companies divide data from the call into ‘packets’ to transmit information from one end to another. During the transit, some of these packets are lost, which creates glitches and other issues with Internet calls as you must have experienced. WaveNetEQ essentially helps Duo create these missing packets of information based on the massive amounts of speech data that Google has. You read the technical nitty-gritties of the technology on the company’s blog.

Google states that 99 percent of Google Duo calls deal with packet losses or network delays. Out of these, 20 percent lose more than 3 percent of the total audio duration while 10 percent of calls lose more than 8 percent of the total duration. That’s a lot of data loss which can prove to be quite problematic at times, especially in the current global scenario.

Explaining the predictive nature of the feature that allows users to keep the conversation going, Google says, “The spectrogram of the past audio signal is used as input for the conditioning network, which extracts limited information about the prosody and textual content. This condensed information is fed to the autoregressive network, which combines it with the audio of the recent past to predict the next sample in the waveform domain.”

“Because Duo calls are end-to-end encrypted, all processing needs to be done on-device,” Google says in the blog post. “The WaveNetEQ model is fast enough to run on a phone, while still providing state-of-the-art audio quality and more natural sounding PLC than other systems currently in use.”

This feature was made available in all Duo calls on Pixel 4 phones with the first Pixel drop in December last year. Google says it is now reaching other models without specifying exactly which. Chances are that other Pixel phones will be first to get it, followed by other devices.

Source Article