Local Voice Assistant Natural Language to SQL (Bachelor Thesis)
📌 Overview
A privacy-focused voice assistant system that operates entirely offline, developed as a Computer Engineering thesis project at Universitat Autònoma de Barcelona. The system enables natural language interaction with databases through:
- Voice command processing via local AI models
- Secure SQL query generation
- Responsive web interface
- Complete data isolation (no cloud dependencies)
🌟 Key Features
| Category | Details |
|---|---|
| Privacy First | Fully offline operation - No data leaves user's device |
| Speech Processing | 95% accuracy STT with WhisperAI + natural-sounding TTS with XTTS v2 |
| NLP to SQL | Context-aware query generation using CodeQwen1.5 7B parameter model |
| User Interface | Responsive web dashboard with query history and voice interaction logs |
🛠 Technical Highlights
- Web Framework: Laravel 10 with Bootstrap 5 frontend
- Speech Core:
- WhisperAI (tiny.en) for efficient local STT
- CoquiTTS XTTS v2 multilingual synthesis
- AI Processing:
- CodeQwen1.5-7B via Ollama LLM server
- Custom prompt engineering for SQL generation
- Database: MySQL with optimized schema for voice interaction metadata
- Performance: less than 5s response time on consumer-grade hardware
🌐 Live Demo
Explore the live implementation of the project:
🔗 https://tfg.netshiba.com/