CSSEL-P2P enables simultaneous speech translation without architectural changes

The article presents CSSEL-P2P, a data-driven approach for simultaneous speech translation that avoids the brittle architectural changes and explicit read/write policies often required by decoder-only LLM systems. The method utilizes fixed-length chunks for cumulative streaming decoding with a rewind-based committed prefix, combined with teacher-labeled prefix-to-prefix targets and bounded waiting for fine-tuning.

• CSSEL-P2P improves streaming quality by +1.54 COMETKiwi over the CSSEL streaming baseline at comparable latency (+0.15s Average Lagging).
• The approach addresses challenges in conversational speech where segmentation boundaries are ambiguous, offering an effective alternative to complex architectural modifications.

This demonstrates that simultaneous speech translation can be achieved effectively through prefix-to-prefix supervision without necessitating structural changes to the model.