BCIwiki (bciwiki.com) — Researchers report that SAGE, a switch-aware EEG-guided soft gating framework for target speaker extraction, achieves 8.67 dB SI-SDR and 88.24% STOI while reducing average switching latency to 2.04 s, according to a preprint posted to arXiv on August 3, 2026 that has not yet undergone peer review.
Under in-trial auditory attention switching, neural noise and intrinsic latency can delay or destabilize attention tracking, and conventional methods struggle with dynamic switches and often cause discontinuities at switching points. SAGE treats in-trial switching as dynamic selection, generating two candidate speech streams with a robust separator and using an EEG-guided switch-aware gating module to produce smooth fusion weights and suppress transition artifacts. The framework integrates latency-compensated alignment and an uncertainty-driven conservative strategy to handle latency discrepancies and fluctuating EEG reliability, outperforming baselines and enabling robust target extraction in dynamic scenarios.