Google develops codec 'Lyra' that realizes high sound quality even at a low bit rate of 3 kbps
On February 25, 2021, Google announced that it has developed a
Google AI Blog: Lyra: A New Very Low-Bitrate Codec for Speech Compression
https://ai.googleblog.com/2021/02/lyra-new-very-low-bitrate-codec-for.html
Most codecs use a model that achieves high sound quality by compressing the audio signal for each sample and transmitting it, but this model cannot reproduce the original sound quality at a low bit rate. However, the model newly developed by Google is said to be able to reproduce voice using the minimum amount of data.
The original sound quality can be reproduced even at a low bit rate, with the limit of 6 kbps of the
The following is the result of the mean opinion score that summarizes the evaluations by multiple subjects, and shows the scores of high-quality voice, Lyra, Opus, and Speex prepared as a reference from the left. In this test, Lyra scored well above Opus with a bitrate of 6kbps.
In fact, you can see how good Lyra is by listening to the samples published on Google's official blog.
First, the following is the original audio.
Next, the voice processed by Opus is as follows. It's not inaudible, but it feels like wireless communication with a walkie-talkie.
And this is the audio processed by Lyra. It's not exactly the same as the original, but you can see that it's not grainy and very clear.
One of the challenges in developing codecs for audio and video is how to improve quality with low data usage. Movies may seem to consume more bandwidth than audio, but in fact the latest video codecs may offer lower bitrates than some audio codecs.
By combining a low bit rate video codec and a voice codec, it is possible that high-quality video calls can be realized even on low-bandwidth networks. The development team says that when Lyra is combined with a video compression codec
Related Posts:
in Software, Posted by log1p_kr