Geometric Iterative Retrieval for Neural Audio Codec Resynthesis
The paper proposes restoring coarse neural audio codec tokens by retrieving through RVQ codebook geometry, not by plain token classification or regression.
The authors use the RVQ layer hierarchy as an iterative path through continuous codebook space. Their method performs contrastive retrieval against that space to resynthesize audio from coarse codec tokens. They test it on speech and music codec restoration tasks and report gains over single-pass token prediction and one-step regression baselines. ArXiv · AI/CL/LG's note
The authors use the RVQ layer hierarchy as an iterative path through continuous codebook space. Their method performs contrastive retrieval against that space to resynthesize audio from coarse codec tokens. They test it on speech and music codec restoration tasks and report gains over single-pass token prediction and one-step regression baselines. ArXiv · AI/CL/LG's note
score 4