Reproduce “CodecFlow: Efficient Bandwidth Extension via Conditional Flow Matching in Neural Codec Latent Space” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bowen Zhang, Junchuan Zhao, Ian McLoughlin, Ye Wang, A S Madhukumar
CodecFlow extends narrowband speech by generating high-resolution neural-codec embeddings conditioned on the input embeddings and frame-level voicing labels, then quantizing and decoding them. In 8 kHz to 44.1 kHz tests, it achieves a high-frequency log-spectral distance of 0.98, versus 1.25 for AP-BWE, and leads overall LSD and LSD-HF, but not every perceptual metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.