Google makes Lyra speech codec open source

Google has released the code of its newly developed speech codec Lyra as beta on Github. The company claims to be able to offer better audio quality with the codec during conversations over the internet.

Google developers explain in a blog post how the Lyra audio codec works: “Every recorded voice message is broken into small pieces every 40 milliseconds, which we call features. These are compressed, converted into data that can be read by a spectrogram and sent to a receiver that uses this data to help reconstruct the sound fragment,” said developers behind Lyra.

The algorithm behind the codec has been trained from thousands of hours of open source audio clips in more than 70 languages. According to Google, this should ensure that this audio codec can be used universally. Google thinks that the codec has many applications. “By donating the code to the open source community, we think this codec can be used in very original ways,” it sounds.

The Lyra codec is designed to work at a bitrate of 3 kbps and, according to the creators behind the compression format, is said to outperform other similar codecs, such as the Opus codec. Google also demonstrates this on its blog. The efficiency of the Lyra codec should help in areas where mobile internet coverage is lower or where internet connections are more heavily loaded.