Abstract
Reports of daily time use are gathered for both large-scale population-level surveys and individual self-tracking in personal informatics systems. However, gathering a complete self-report record of an individual’s daily activities is time consuming and cognitively demanding. Gathering structured time records via spoken narrative can reduce the burden of self-tracking and support natural “storytelling’’ as a method of data collection. We designed, built, and evaluated TellTime, a voice interface to a calendar using a large language model. Over three days, 18 participants completed the Day Reconstruction Method with three different calendar interfaces: manual-only, voice-only, and hybrid. We analyzed the user experience with surveys, semi-structured interviews, and telemetry data. Our findings show that users prefer hybrid interaction, where they first gather data via a spoken narrative, and then manually make precise adjustments to specific activities. This work suggests that LLMs can enable voice interaction with calendar systems, and that spoken narrative can improve the user experience of gathering self-report time records.