Google hopes to end low-quality video calls by using artificial intelligence to fill audio gaps caused by bad connections.
As the BBC reports, “WaveNetEQ” works by using a “library” of speech data to fill in short discussion sections in a realistic/plausible way. Artificial intelligence is trained to produce mainly syllable sounds and can fill gaps of up to 120 milliseconds.
This development takes place as the use of video calls becomes increasingly important, given the crisis of the corona-virus pandemic.
When an Internet call occurs, the data are divided into small chunks, the “packets”. A bad connection can mean that these “packets” arrive at their destination in the wrong order and at the wrong time, or are completely lost. This means poor call quality.
According to Google, 99% of calls made through its “Duo” app experience some sort of sound problem. Of these calls, 20% lose more than 3% of their total sound, while 10% lose almost 1/10.
“WaveNetHQ” works by creating specific speech data to fill the gaps created by audio losses.
Artificial intelligence was trained using the voices of 100 people in 48 languages to be able to “learn” the general characteristics of a human language, regardless of dialect.
According to the BBC, the system is available on Google’s Pixel 4 smartphone, with the company planning to switch to more Android devices later this year.