FADHILI-Josue/Talk2AI-kinyarwanda

# Kinyarwanda Voice Assistant 🤖 An intelligent voice assistant that supports natural conversation in Kinyarwanda, built using speech recognition, fuzzy NLP matching, and text-to-speech. Created for the Intelligent Robotics course. ## 🔑 Key Features - 🎙️ Speech-to-text using **KinyaWhisper** - 🧠 Context-aware response generation via **fuzzy logic** - 📢 Kinyarwanda **text-to-speech** - 🔇 Advanced **noise reduction** - 🗣️ Voice Activity Detection (VAD) - 📊 Conversation analytics and web interface via **Gradio** ## 🛠️ Tech Stack - **AI & NLP**: Hugging Face, FuzzyWuzzy, Python-Levenshtein - **Audio**: Librosa, Soundfile, WebRTC VAD - **Interface**: Gradio - **Optimization**: Noisereduce, FFmpeg ## 🚀 Quick Start ```bash git clone github.com cd Talk2AI-kinyarwanda python -m venv .venv source .venv/bin/activate # or .\.venv\Scripts\activate on Windows pip install -r requirements.txt copy .env.example .env # add you token from hugging face python main.py ``` Then open localhost ## ⚙️ Configuration - Customize questions/responses in `nlp_mapping.json` - Sample recordings in `/sample_inputs` (supports WAV) ## 💡 Interface Usage - Record/upload audio - Click **Submit** - Listen to the reply & view history (transcription, matched Q&A) ## 📋 Example | Input | Matched Q | Response | |----------------|------------------|--------------------------------------------| | muraho | muraho? | Muraho! Ndi hano kugufasha. Ni iki wakwifuza kumenya? | | wakozwe na nde | wakozwe na nde? | Nakozwe na Fadhili Josue mukigo cya Rwanda Coding academy. |