arXiv:2610.32702v1 Announce Type: new Abstract: Full-duplex speech models can listen and speak at the same time, enabling natural, low-latency interaction. In counseling, however, these models often focus on the latest utterance and lose track of the seeker's underlying problem. We call this failure Stressor Blindness. We propose CounselPlex, a stressor-aware full-duplex framework that uses two operations. CounselPlex updates a short stressor state during seeker silence and injects this state into positions while listening. We also introduce RoPE Position Shift Augmentation (RPSA), which broadens positional coverage during training and improves performance at longer turn depths in our evaluation. LLM-judge evaluations show improved counseling quality and stressor grounding over strong baselines.
CounselPlex: A Stressor-Aware Full-Duplex Counseling Model
A new arXiv paper (2610.32702v1) proposes CounselPlex, a stressor-aware full-duplex framework for counseling speech models that addresses what its authors call Stressor Blindness — the tendency of full-duplex models to focus on the latest utterance and lose track of the seeker's underlying problem. CounselPlex updates a short stressor state during seeker silence and injects that state into positions while listening, and the authors introduce RoPE Position Shift Augmentation (RPSA) to broaden positional coverage during training and improve performance at longer turn depths. LLM-judge evaluations show improved counseling quality and stressor grounding over strong baselines.
Run your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.