The next-generation audio codec Encodec uses a sophisticated neural network architecture to pack a lot of audio information into a little amount of storage.
The codec might be useful for optimizing cell phone calls and Metaverse experiences.
The MP3 codec became the de-facto standard for transferring audio files on the internet during the Nineties and beyond due to its great efficiency and integrated support by notable products like the enduring Winamp player.
Currently, a new codec aims to once more change the course of history by providing even greater efficiencies and bandwidth savings. The key is an artificial intelligence program that can “hypercompress” audio streams.
Encodec was imagined by meta researchers as a possible means of sustaining “present and future” high-quality experiences in the metaverse.
A neural network that has been trained to “push the frontiers of what is achievable” in audio compression for online applications is the new technology.
In comparison to the MP3 standard, the device may achieve “an approximate 10x compression rate.”
The AI was “end to end” meta-trained to reach a particular target size after compression.
Encodec can reduce a 64 Kbps MP3 data stream to 6 Kbps, requiring only 6,144 bytes (yep, bytes) to maintain the original’s quality.
An industry first, according to the researchers, the codec can compress speech-related 48 kHz stereo audio samples.
According to the sample posted on Meta’s AI blog, the AI-based method can “compress and decompress audio in real-time to state-of-the-art size reductions,” with potentially amazing outcomes.
With the use of psychoacoustics, traditional codecs like MP3, Opus, or EVS divide the signal into its component frequencies (the study of human sound perception).
The methods used by Encodec are based on a sophisticated architecture that consists of three components: an encoder, a quantizer, and a decoder.
According to Meta, Encodec still has a lot of room for advancement for even lower file sizes and is capable of encoding and decoding audio data in real-time on a single CPU core.
The new approach may be able to ensure higher-quality phone conversations in regions with less than ideal cell coverage, in addition to supporting next-gen Metaverse experiences on current internet connections.
To read our blog on “The Cloud City Metaverse, Pakistan’s first massive web3 project, is about to launch,” click here.
