Descript Audio Codec

The Descript Audio Codec (DAC) is a high-fidelity general neural audio codec introduced in the paper 'High-Fidelity Audio Compression with Improved RVQGAN'. It compresses 44.1 kHz audio into discrete codes at just 8 kbps — roughly 90x compression — while preserving quality and minimizing artifacts.\n\nA single universal model handles speech, environmental sounds and music, making it a drop-in replacement for EnCodec in audio language modeling applications such as AudioLMs, MusicLMs and MusicGen. Training and inference code plus pretrained weights are published on GitHub.\n\nDAC's combination of high sample rate, low bitrate and broad-domain coverage has made it a standard tokenization layer for generative audio research.