Communication barriers between deaf individuals and the general public remain a persistent challenge in Arabic-speaking communities. This paper presents Gesture to Voice, a real-time Arabic Sign Language (ArSL) recognition system that translates hand gestures into spoken Saudi Arabian audio output. MediaPipe extracts 21 hand landmarks per hand (126 features total), and a Long Short-Term Memory (LSTM) neural network processes temporal sequences of 30 consecutive frames to classify gestures. A two-stage prediction stabilization mechanism combining confidence thresholding and majority voting ensures reliable output. The system achieves a best validation accuracy of 84.26% and reliable real-time performance across three gesture classes. Unlike prior work producing text-only output, Gesture to Voice uniquely delivers spoken SaudiArabianc responses, addressing a critical gap in localized assistive technology.