tchamna/african-text-to-speech

# AfricanVoice: AI-Powered Multilingual Voice Translator ## 🚀 Project Overview **AfricanVoice** is a sophisticated AI-driven translation application designed to bridge the linguistic gap between **French, English, and Spanish** and **Nufi** (a Bamileke language from Cameroon). Unlike traditional dictionary lookups, this project leverages state-of-the-art **Machine Learning** and **Natural Language Processing (NLP)** techniques to understand spoken language and retrieve semantically similar translations, even when the phrasing differs from the database. This project serves as a demonstration of applying modern AI to low-resource language preservation and education. ## 🌐 Live Demo Try the live version of AfricanVoice deployed on Azure: **african-text-to-speech-ehaj… _Figure: Top — main UI (record & results). Bottom — semantic matches and Nufi playback._ No installation required - just click the link and start speaking! ## 🧠 Key AI & Machine Learning Features ### 1. Advanced Speech Recognition (ASR) - **Engine:** OpenAI Whisper (Large-v3 model). - **Capability:** Transcribes spoken French, English, and Spanish with high accuracy, handling accents and background noise effectively. - **Multi-Model Architecture:** Includes a development mode to compare performance across Whisper `Small`, `Medium`, and `Large-v3` models in real-time. ### 2. Semantic Search & Vector Embeddings - **Embeddings:** Uses `SentenceTransformers` (`paraphrase-multilingual-MiniLM-L12-v2`) to convert text (French, English, Spanish) into high-dimensional vector representations. - **Vector Database:** Implements **FAISS (Facebook AI Similarity Search)** for ultra-fast similarity retrieval. - **Benefit:** Allows the system to understand *meaning* rather than just keywords. - *Input:* "Je me sens bien" (I feel good in French) - *Match:* "Je suis à l'aise" (I am at ease) -> **90.5% Similarity** - *Input:* "I feel good" (English) …