Meta has announced its AI-powered audio compression method, named EnCodec. The claim is that rather than being another means of compressing audio to save digital bandwidth that this is a next-generation system that can tackle increasing demands by improving sound quality on low-bandwidth connections.
In other words using AI on poor quality digitally compressed or interrupted signals in order to put back the data that’s missing.
Meta states that research has shown that AI could be used to help situations such as poor audio quality or stalling of audio messages in low connectivity areas. Meta’s three-part system is trained to compress audio to the desired size. According to research, EnCodec can achieve an approximate 10x compression rate compared with MP3 at 64kbps, all without a loss in quality.
Better audio experiences, regardless of the location
The three-part system consists of the encoder part, which takes uncompressed data and then transforms it into a higher dimensional but lower frame rate. The quantizer is the part that allows for targeted sizes to be met. Finally, the decoder is then used to turn the compressed signal back into a waveform that resembles the original as closely as possible.
Such advances could see the improvement of not only audio messages but music, streaming movies, video calls, and playing games in VR. Meta notes that currently a stable and quick internet connection is needed for many media experiences, such as the metaverse, and to deliver these experiences to a wider audience uninterrupted, future compression techniques will need to overcome current limitations.