oqaidik/morocco-dialect-ai-assistant

🇲🇦 Morocco Voice MVP — Multilingual Speech AI System (EN ↔ Darija) Morocco Voice is a production-ready multilingual speech AI pipeline that enables bidirectional voice and text communication between English and Moroccan Darija, a low-resource dialect. 🔗 Live Demo (Mobile-friendly): mvp-morocco-voice.onrender.… 📘 API Documentation (Swagger): mvp-morocco-voice.onrender.… 🚀 What This Project Demonstrates This MVP is designed as a real-world AI system, not just a demo. It showcases: End-to-end Speech AI pipelines Integration of STT → LLM → TTS Clean, stateless FastAPI microservice Deployment-ready architecture for mobile apps AI-assisted communication for a low-resource language (Darija) 🎯 Core Capabilities 🔊 English → Darija (“Speak for Me”) Input English text Output Darija text (Arabic script) Darija phonetic transcription Darija speech audio (TTS) Context-aware alternative replies Use case Helping non-Darija speakers communicate naturally in Morocco (travel, services, daily life). 🎧 Darija → English (“Translate”) Input Darija speech audio Output English transcription English translation English speech audio (TTS) Use case Voice-based translation for real-life conversations. 🧠 AI System Architecture Client (Mobile / Web Demo) | v FastAPI Backend (Stateless) ├── Speech-to-Text (STT) │ └── Audio normalization & transcription ├── Large Language Model (LLM) │ └── Translation, reasoning, Darija generation ├── Text-to-Speech (TTS) │ └── Audio synthesis (Darija / English) └── JSON API (Mobile-ready) Design choices Modular services (STT / LLM / TTS decoupled) JSON-first LLM prompting (robust parsing) Explicit latency tracking Graceful error handling for STT & TTS failures 🛠 Tech Stack Backend Python, FastAPI Async request handling Speech & Language Speech-to-Text: Whisper-compatible STT LLM: OpenAI (structured JSON prompting) Text-to-Speech: Google Cloud Text-to-Speech Deployment Render (production) …