marknature/Sahwira_Inwa

Sahwira Inzwa is an AI-driven application designed to translate sign language gestures into audible Shona speech in real time. # Sahwira Inzwa Bridging the communication gap between deaf individuals and non-signers through AI-powered Shona translation. ## 🌟 Project Overview **Sahwira Inzwa** is an AI-driven application designed to translate sign language gestures into **audible Shona speech** in real time. By leveraging computer vision and natural language processing, the system enables inclusive, natural communication, reducing the high costs and limited availability of human interpreters in Zimbabwe. Watch demo ## 🚀 Key Features * **Real-time Gesture Recognition:** Uses computer vision to detect and interpret hand movements instantly. * **Localized Audio Output:** Specifically tailored to produce natural Shona voice output, a language often overlooked by global English-centric technologies. * **Hand Landmark Tracking:** Utilizes **Google MediaPipe** to track 21 specific hand joints and fingertips for high precision. * **Sentence Logic:** Includes intelligent processing that adds characters to a sentence only when a gesture is held steadily for 2 seconds, with support for spaces and deletions. * **Intuitive UI:** Designed with accessibility in mind, featuring a large camera preview and simple playback controls. ## 🛠️ Tech Stack * **Artificial Intelligence:** Scikit-learn (Random Forest Classifier), TensorFlow/PyTorch. * **Computer Vision:** MediaPipe, OpenCV. * **Backend:** Python, Flask. * **Frontend:** Next.js (App Router), React, TypeScript, Tailwind CSS. * **Speech Technology:** Google Cloud Speech API, Shona Text-to-Speech (TTS), gTTS. ## 📂 System Architecture The system follows a streamlined pipeline to ensure low latency (<2 seconds): 1. **Input:** Live video feed captured via smartphone or laptop camera. 2. **Detection:** MediaPipe identifies 21 landmarks and extracts a 42-dimension feature vector. 3. **Classification:** A trained ML model predicts the gesture meaning. 4. **Translation:** Text is mapped to its Shona equivalent. 5. **Synthesi …