Synthia: Voice AI for Dementia Care
Real-time voice AI assistant for memory care. Whisper STT, ElevenLabs TTS, GPT-4 context engine.
Overview
Synthia is a real-time voice AI healthcare assistant purpose-built for dementia care support. The system chains OpenAI Whisper for speech recognition, GPT-4o for contextual understanding, and ElevenLabs for natural voice synthesis into a seamless conversation pipeline. Healthcare-specific prompt engineering ensures responses are empathetic, clear, and clinically appropriate — avoiding the ambiguity that can distress dementia patients.
Challenge
Voice AI in regulated healthcare requires more than accuracy: latency, affect, and failure-mode design all become patient safety issues. The challenge was to build a reliable real-time pipeline that could handle audio processing, model inference, and voice synthesis end-to-end with low enough latency to feel conversational.
Outcomes
- End-to-end voice pipeline: Whisper → GPT-4o → ElevenLabs under 2s latency
- Healthcare-specific system prompt with dementia care conversational protocols
- Web Audio API integration for real-time audio capture and playback
- Modular service architecture separating STT, LLM, and TTS concerns
- Loom video walkthrough demonstrating live conversation flows