Talking Spell: A Wearable System Enabling Real-Time Anthropomorphic Voice Interaction with Everyday Objects

Abstract: Virtual assistants (VAs) have become ubiquitous in daily life, integrated into smartphones and smart devices, sparking interest in AI companions that enhance user experiences and foster emotional connections. However, existing companions are often embedded in specific objects-such as glasses, home assistants, or dolls-requiring users to form emotional bonds with unfamiliar items, which can lead to reduced engagement and feelings of detachment. To address this, we introduce Talking Spell, a wearable system that empowers users to imbue any everyday object with speech and anthropomorphic personas through a user-centric radiative network. Leveraging advanced computer vision (e.g., YOLOv11 for object detection), large vision-LLMs (e.g., QWEN-VL for persona generation), speech-to-text and text-to-speech technologies, Talking Spell guides users through three stages of emotional connection: acquaintance, familiarization, and bonding. We validated our system through a user study involving 12 participants, utilizing Talking Spell to explore four interaction intentions: entertainment, companionship, utility, and creativity. The results demonstrate its effectiveness in fostering meaningful interactions and emotional significance with everyday objects. Our findings indicate that Talking Spell creates engaging and personalized experiences, as demonstrated through various devices, ranging from accessories to essential wearables.