Reproduce “DIAL-S2ST: Learning Latency Control for Simultaneous Speech-to-Speech Translation” 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
Jeongsoo Choi, Junseok Ahn, Joon Son Chung
DIAL-S2ST adapts Hibiki for adjustable-latency French-to-English speech translation. A frozen teacher provides lookahead predictions while a student with a trainable LoRA adapter learns from its own generated histories; changing the adapter scale at inference adjusts emission timing. On CVSS at 1.8 seconds LAAL, it scores 32.5 ASR-BLEU, ahead of two alternatives, in the evaluated Hibiki setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.