Voice-Based Virtual Assistant for Emotionally Sensitive Interactions

This paper presents Luna, an emotionally intelligent, voice-based virtual assistant built to engage users in meaningful, empathetic conversation. Unlike conventional AI assistants that emphasize efficiency and task execution, Luna is designed to understand human emotions and respond with care, warmth, and contextual understanding. Luna integrates local large language models (LLMs) using llama-cpp-python, speech recognition via OpenAI’s Whisper, and expressive voice output via Coqui’s YourTTS. The assistant interacts through a chat-like GUI designed using Tkinter, offering both voice and text input. The system maintains recent dialogue history and adapts its responses with emotional sensitivity based on user input. This paper discusses the architecture and technical implementation of Luna, demonstrating how emotion-aware AI can enhance digital companionship in an entirely offline environment.