AI-Based Glove for Hearing Impaired

Abstract: The hearing-impaired community faces persistent communication barriers due to limited sign language fluency among the general population. Existing assistive solutions are predominantly camera-dependent, cost- prohibitive, or restricted to single-language support. This paper presents a low-cost, real-time assistive glove system that translates American Sign Language (ASL) gestures into speech with multilingual output capabilities. The proposed system employs five flex sensors (36 Hz sampling) integrated with an MPU6050 inertial measurement unit to capture temporal signing patterns. A quantized Bidi rectional Long Short-Term Memory (Bi-LSTM) model performs on-glove inference for low-latency recognition, achieving 92% accuracy across 36 gesture classes (A-Z, 1-10). Recognized English tokens are translated to Kannada, Hindi, French, and Spanish via cloud APIs, with text-to-speech output. Bidirectional interaction is enabled through speech capture, conversion to text, and display on an integrated OLED screen. The system emphasizes portability, affordability, and robustness for daily use, with experimental validation demonstrating conversational latency and day-long battery feasibility. This work contributes a comprehensive framework for inclusive communication in multilingual environments.

Keywords: Sign language recognition, assistive technology, Bi-LSTM, wearable sensors, multilingual translation,bidirectional communication, embedded machine learning, ESP32-S3.