SwitchCodec: Adaptive Residual-Expert Sparse Quantization for High-Fidelity Neural Audio Coding

Xiangbo Wang, Wenbin Jiang, Jin Wang, Yubo You, Sheng Fang, Fei Wen · 2026

Read SwitchCodec: Adaptive Residual-Expert Sparse Quantization for High-Fidelity Neural Audio Coding by Xiangbo Wang, Wenbin Jiang, Jin Wang, Yubo You, Sheng Fang, Fei Wen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-20362.txt.

Description

Xiangbo Wang and five colleagues present SwitchCodec, a neural audio codec built for high-fidelity compression. Neural codecs often use residual vector quantization, stacking codebooks to describe each frame of sound. A fixed number of codebooks fits all content alike, which is wasteful for simple sounds and too thin for complex ones.

SwitchCodec uses residual experts vector quantization, which switches the coding effort to match the content. Simple frames can use fewer resources while hard frames get more. Adaptive coding of this kind could give streaming and speech systems better sound at the same data rate.

Catalog record

CREATORXiangbo Wang, Wenbin Jiang, Jin Wang, Yubo You, Sheng Fang, Fei Wen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-20362

Browse the full library · About the Tower of Babel