StrokeVoice: A Low-Cost Hybrid Multimodal AAC Ecosystem for Post-Stroke Aphasia and Dysarthria

We present StrokeVoice, a low-cost, hybrid multimodal Augmentative and Alternative Communication (AAC) ecosystem designed as a Silent Speech Interface (SSI) for patients with post-stroke aphasia and dysarthria. The system integrates a throat vibration sensor with an acoustic amplification module to capture subtle, non-audible articulatory movements during silent speech. The acquired signals are digitized and transmitted via Bluetooth Low Energy (BLE) to a mobile application, where a lightweight deep learning model performs real-time classification. To enhance communicative utility, we propose a conceptual framework utilizing a Large Language Model (LLM) to reconstruct coherent, full sentences from the discrete, recognized keywords. Experimental evaluations across five subjects demonstrate that StrokeVoice restores intelligible speech states with a system latency of less than 1 second, establishing its viability as a portable, affordable communication aid in both quiet and noisy environments.