EnCodec is a state-of-the-art neural audio codec developed by Meta AI and released as an open-source project on GitHub, where it has gathered roughly 4,100 stars. It uses deep learning to compress audio far more aggressively than traditional codecs while keeping perceptual quality high, supporting mono audio at 24 kHz and stereo audio at 48 kHz.
EnCodec is built around a convolutional encoder-decoder trained with adversarial and reconstruction losses, combined with a transformer-based language model over the quantized audio tokens. This design made it a foundational building block for the recent wave of audio generation research: models such as MusicGen and AudioGen build directly on EnCodec tokens to generate music and sound from text. The repository ships pretrained models, training code and simple command-line tools for compressing and decompressing audio files. The project was last actively updated in 2024 and remains a reference implementation for learned audio compression.
EnCodec is completely free and open source under the MIT license. There are no paid tiers, hosted versions or usage limits; you download the code and pretrained weights and run everything on your own hardware.
EnCodec is for audio researchers, ML engineers building text-to-audio or speech generation systems, and developers who need a modern learned codec as a component in a larger media pipeline.