Abstract:
Early Retirement: An Open-World Game with Live NPC Interaction presents an open-world PC game developed in Unity, featuring real-time, voice-activated NPC interactions powered by a fine-tuned compact language model (SmolLM2-360M), which explores themes of trauma and self-discovery, aiming to enhance immersion by replacing static dialogue trees with dynamic, AI-driven conversations; this project specifically investigates whether these real-time, voice-driven AI interactions can significantly enhance narrative immersion and player agency in such open world games. The system utilizes speech processing APIs alongside a custom backend to enable spontaneous, emotionally aware NPC responses to live player speech, a feat achieved through efficient fine-tuning of the language model using a substantial domain-specific dataset of emotional dialogue scenarios, ensuring functionality even under resource constraints. With an average response latency of approximately 3 seconds, the system demonstrates practical viability despite occasional incoherent outputs, and preliminary user testing indicated increased engagement and perceived realism, although further comprehensive evaluation is needed to fully assess long-term usability and sustained coherence. This work therefore contributes a novel framework for integrating lightweight language models into real-time game environments, enabling responsive, emotionally aware NPCs without requiring high-end hardware, thereby showcasing the transformative potential of AI in interactive media and providing valuable insights into the unique challenges and innovative solutions encountered during its development.
