Reproduce “RubatoCodec: Budgeted Adaptive Speech Tokenization with Streaming Decoding” 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
Chihiro Arata, Shohei Taniguchi, Gouki Minegishi, Kan Hatakeyama-Sato
RubatoCodec represents speech with a fixed event budget and content-adaptive event placement, encoding discrete codes with durations on a 50-Hz grid. This keeps sequence length predictable while allowing events to span different durations. In an English placement ablation, adaptive placement improves depth-16 dWER and PESQ-WB modestly, but uniform placement performs better on depth-8 dWER; results do not establish universal superiority.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.