Nov 02, 2022 08:00:00

'Encodec' realizes compression rate and compression speed exceeding conventional compression with audio compression using AI

The research team of Meta AI by Gabriel Sinaave and others has announced a study that AI can be used for '

compression ' of voice on the Internet to achieve further compression than conventional compression. It explains that compression using AI allows you to enjoy a rich multimedia experience.

Using AI to compress audio files for quick and easy sharing
https://ai.facebook.com/blog/ai-powered-audio-compression-technique/

'Compression' is an integral part of today's Internet, and compression enables high-quality images and streaming. However, current compression technology requires a fast internet connection and ample storage space, making a high-quality, uninterrupted internet experience available only to a select few.

Therefore, the research team of Meta AI is studying compressing audio data using AI. He announced that an AI-based approach can compress and decompress audio in real time and achieve state-of-the-art size reduction. In this research, compared to MP3 of 64kbps, about 10 times the compression rate has been put into practical use with stereo sound of 48kHz sampling, which is CD quality, without quality loss.

Compression using AI called 'Encodec', which was learned and created by Meta AI, consists of three parts.

◆1: Encoder
Converts uncompressed data into a higher-dimensional low-frame representation than before.

◆2: Quantizer
Compresses the data received from the encoder to the target size. Encodec is trained to output the desired size while retaining the most important information to reconstruct the original signal.

◆3: Decoder
It restores the signal compressed by the quantizer to a waveform that is as close as possible to the original signal. Encodec identifies changes that humans cannot perceive and enables lossy compression at low bit rates.

By compressing audio using Encodec, we have achieved state-of-the-art results in low-bit-rate audio compression from 1.5 kbps to 12 kbps, and have announced that real-time audio encoding and decoding are possible with a single CPU core.

In future research, we plan to compress the audio to even smaller file sizes without significantly degrading the quality, and to explore spatial audio compression as well. In addition, we plan to work on compression research using AI in the field of video, and as a result of the research, 'the possibility that people around the world will have a richer and faster online experience regardless of the speed of the Internet connection. there is,” Gabriel Sinaave and others explain.