An interactive codec pipeline comparing vector quantization with residual vector quantization.

Compress sound into codec tokens

An audio codec compresses and rebuilds sound. Its encoder makes a short sound description, its quantizer chooses numbered codewords, and its decoder turns those choices back into audio.

Vector quantization (VQ) chooses one nearest codeword. Residual vector quantization (RVQ) lets later codebooks encode the residual—the detail earlier choices missed. A frame is a short sound slice; a bit is one 0-or-1 storage unit. Eight 1,024-choice codebooks use 80 bits/frame and produce 600 tokens/s at 75 frames/s.