descriptinc/descript-audio-codec

Health v2 · maintenance only

Adoption not part of the score

1846 stars · 187 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Descript Audio Codec (.dac) is a high-fidelity neural audio codec based on improved RVQGAN that compresses audio at roughly 90x compression (e.g., 44.1 kHz audio down to 8 kbps). It ships as a pip-installable Python package with CLI encode/decode commands and pretrained weights for 16, 24, and 44.1 kHz mono/stereo audio.

Use cases

• compress audio files to very low bitrates with high fidelity
• encode audio into discrete codes for audio language models like AudioLM or MusicGen
• replace EnCodec as a tokenizer in generative audio pipelines
• decode .dac files back to wav audio
• preprocess speech, music, and environmental audio for machine learning

When to choose

• you need a drop-in EnCodec replacement with better compression at 44.1 kHz
• you need discrete audio tokens for training audio generation models
• you want MIT-licensed codec weights for speech, music, and general audio

When to avoid

• you need real-time streaming or ultra-low-latency encoding
• you need multichannel (beyond stereo) or video codecs
• you cannot run GPU/CPU inference with PyTorch dependencies

Facets

library · maturity stable

audio-processing compression machine-learning deep-learning machine-learning media python cross-platform cli audio-codec neural-codec rvqgan pytorch audio-compression encodec-alternative audio

Member repositories

For agents

markdown · JSON · MCP: product_card(name="descriptinc/descript-audio-codec")