BYUMVUHOREAimable/kinyarwanda-voice-assistant

Kinya is a cutting-edge voice assistant designed to understand and respond naturally in Kinyarwanda. Powered by advanced speech recognition and synthesis, it bridges the gap between users and technology through seamless voice interactions. # Kinya Voice Assistant 🎙️ A smart, conversational voice assistant for Kinyarwanda speakers Kinya is a cutting-edge voice assistant designed to understand and respond naturally in Kinyarwanda. Powered by advanced speech recognition and synthesis, it bridges the gap between users and technology through seamless voice interactions. ## ✨ Key Features - 🗣️ **Native Kinyarwanda Support** – Built specifically for Rwandan language and accents - 🎙️ **Speech-to-Text** – Accurate voice transcription using NeMo AI models - 🧠 **Smart Responses** – Combines rule-based logic + ChatGPT for natural conversations - 🔊 **Text-to-Speech** – Fluent Kinyarwanda audio output with MB-iSTFT-VITS2 - 🌐 **Web Interface** – Easy-to-use Gradio UI for instant testing ## 🏁 Quick Start 1. Clone the repository: ```bash git clone github.com cd kinyarwanda-voice-assistant ``` ## 🚀 How to Test the Project ### Prerequisites 1. **Google Drive Setup** - Upload the model files to your Google Drive - Access the model files here: TTS Model (Google Drive) 2. **Hugging Face Account** - Create an account on Hugging Face - Copy your API key 3. **Google Colab Setup** - Open Google Colab - Add your Hugging Face token in Colab secrets: - Name: `HF_TOKEN` - Value: Your Hugging Face API key ### Installation Steps 1. Upload `kin_assistant.ipynb` to your Colab environment 2. Install dependencies (see Dependencies section below) 3. Run each code cell sequentially 4. Launch the Gradio interface to start chatting! ## 🔍 How It Works Kinya processes conversations in 3 steps: 1. **Listen** → Converts speech to text (STT) 2. **Understand** → Analyzes queries with hybrid NLP 3. **Respond** → Generates spoken replies (TTS) ## ⚙️ Tech Stack ### Speech Recognition - **Framework**: NVIDIA NeMo - **Model**: Pretrained Kinyarwanda STT from RW-DEEPSPEECH-API - **Input**: Handles mic/web audio with noise reduction ### Natural Language Processing **Hybrid Engine …