snac
SNAC is an innovative audio codec that compresses audio into hierarchical tokens for low-bit-rate efficiency. Its applications greatly benefit language models in audio generation. SNAC provides pre-trained models for mono audio across various bitrates and sample rates, suitable for both music and speech. It economizes on bitrate by sampling coarse tokens less frequently, making it ideal for modeling audio structures up to 3 minutes long. Implement SNAC in Python to explore practical encoding and decoding solutions.