Reproduce “GS-Codec: A Gaussian-Splatting Bottleneck for Neural Audio Coding” 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
Ron Aluf, Alon Canfi, Eliya Nachmani
GS-Codec compresses speech by fitting an encoder latent with weighted Gaussian primitives, then transmitting quantized parameters. Users adjust rate after training through primitive count and parameter bit depth, though both settings are discrete. On a matched LibriSpeech test-clean comparison near 6 kbps, it scores higher in UTMOS and speaker similarity than RVQ and FSQ; the result is limited to that English-speech test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.