Abstract
Speech-impaired individuals rely on sign language as their primary communication mode; however, the vast majority of the general population cannot understand it. Hand2Hear is a wearable smart glove that bridges this gap by translating Indian Sign Language (ISL) hand gestures into intelligible text output in near real time. The system employs five resistive flex sensors to measure finger flexion and an inertial measurement unit (IMU) integrated on an Arduino Nano 33 BLE Sense Rev2 microcontroller for orientation capture. A count-based, per-sensor threshold classification algorithm is implemented on-device, reliably detecting a vocabulary of twelve single-finger and two-finger combination gestures mapped to high-frequency ISL communication words. A structured multi-session evaluation across three participants (360 total trials) yields an aggregate accuracy of 89.2%, with per-gesture F1-scores ranging from 0.83 to 0.93. End-to-end gesture-to-label latency is measured at 13–18 ms. Per-sensor calibration data confirms an inverse sensor response across all five fingers, with individual threshold midpoints ranging from 60 to 260 ADC counts. The implemented architecture, including the count-based mutual-exclusivity constraint and EMA-based noise filtering, serves as the validated foundation for future integration of TinyML inference, I2S audio output, and a complete glove assembly.