An advanced voice assistant that comprehends and communicates in Kinyarwanda, delivering a smooth conversational experience through speech recognition and synthesis
# Kinya Voice Assistant 🎙️
An advanced voice assistant that comprehends and communicates in Kinyarwanda, delivering a smooth conversational experience through speech recognition and synthesis.
✨ Highlights
Voice Command Recognition: Captures and interprets Kinyarwanda voice inputs
Speech-to-Text (STT): Precise conversion of Kinyarwanda speech to text using NeMo models
Language Processing: Processes queries with a blend of rule-based logic and ChatGPT integration
Text-to-Speech (TTS): Produces clear, natural Kinyarwanda audio using MB-iSTFT-VITS2
Engaging Interface: Intuitive Gradio web interface for seamless user interaction
📖 Overview
KinyaVoice Companion is a groundbreaking initiative that enables effortless voice-based interaction in Kinyarwanda. The system operates in three phases:
Voice Recognition: Transforms spoken Kinyarwanda into text
Query Analysis: Evaluates the text and crafts suitable responses
Voice Synthesis: Converts responses into fluent Kinyarwanda speech
🎥 Demonstration
Command-Line Execution
Web Interface
Interaction Sample
🔨 System Architecture
Speech-to-Text (STT)
Developed with NVIDIA NeMo framework
Leverages a pretrained Kinyarwanda model from RW-DEEPSPEECH-API
Processes diverse audio inputs into text format
Language Processing
Employs a dual approach:
Pattern-based rules for standard queries
ChatGPT integration for complex inquiries
Custom intent detection mechanism
Facilitates various dialogue types (greetings, inquiries, casual conversation)
Text-to-Speech (TTS)
Built on KinyaTTS
Utilizes MB-iSTFT-VITS2 model architecture
Produces authentic Kinyarwanda speech output
🛠 Installation Guide
Requirements
Python 3.x
NVIDIA GPU (preferred for best performance)
Google Colab (for executing the notebook)
Dependencies
pip install -e /path/to/Inference/ # KinyaTTS
pip install "numpy 1.26.0"
pip install Cython
pip install gradio
pip install openai # …