Sign language serves as a primary communication medium for deaf and hard-of-hearing individuals. However, the lack of widespread sign language literacy among the general population creates significant communication barriers in public service environments such as hospitals, schools, banks, and administrative offices. This paper presents a lightweight Edge–Cloud architecture for real-time Vietnamese Sign Language (VSL) translation into Vietnamese speech using an embedded Raspberry Pi platform. The proposed system integrates a YOLO-based hand detection module, a convolutional neural network (CNN) for static hand-sign classification, cloud-assisted natural language processing for sentence reconstruction, and text-to-speech synthesis for voice generation. The work desmontrate a temporal confirmation mechanism and automatic phrases termination strategy to reduce classification noise and eliminate manual input controls, as well as contribute a curated dataset of 7070 Vietnamese hand-signalphabet gestures. Experimental results demonstrate a classification accuracy upto 98.5%, while maintaining real-time operation on a resource-constrained embedded device. The proposed system offers a complete, low-cost and deployable end-to-end embedded sign-to-speech prototype solution for assistive communication and inclusive smart services. It is especially useful and convinient for disability to communicate and express their thoughts directly with others in public services, hospitals, banks or edge for smart customer services.

Vietnamese Sign Language, Edge AI, Computer Vision, YOLO, CNN, Raspberry Pi, Assistive Technology, Human-Centered AI