Virtual Speech Therapist: A Clinician-in-the-Loop AI Speech Therapy Agent for Personalized and Supervised Therapy
Multi-agent LLM system drafts personalized plans, reviewed by real clinicians for safety.
Virtual Speech Therapist (VST), introduced in a May 2026 preprint by Shakeel Sheikh and colleagues, is an intelligent agent-based platform designed to streamline stuttering assessment and therapy planning. VST integrates state-of-the-art deep learning for robust classification of stuttering types from patient speech samples. It then initiates an agentic reasoning process where specialized LLM agents autonomously generate, critique, and iteratively refine individualized therapy plans. A dedicated critic agent evaluates these plans for clinical safety, methodological soundness, and alignment with peer-reviewed guidelines. The final draft is presented to a clinician, who provides feedback and approves a finalized plan—maintaining a clinician-in-the-loop paradigm throughout.
Experimental evaluation by expert speech therapists confirmed that VST consistently produces high-quality, evidence-based therapy recommendations. The system's interactive user interface is available online for real-time assessment and planning. While still under review, VST demonstrates strong potential to augment clinical workflows, reduce clinician burnout, and improve therapeutic outcomes for individuals with speech impairments—especially critical given the growing demand for accessible, personalized speech therapy services.
- Uses deep learning classification to identify stuttering types from patient speech samples.
- Multi-agent LLM system autonomously generates, critiques, and refines therapy plans with a dedicated safety critic.
- Clinician-in-the-loop paradigm ensures all plans are reviewed and approved by a human expert before patient delivery.
Why It Matters
AI-assisted speech therapy could reduce clinician workload while delivering personalized, evidence-based care at scale.