AudiLens: Configurable LLM-Generated Audiences for Public Speech Practice

Abstract

AudiLens is a large-language model (LLM)-based audience simulator for public speech practice that allows speakers to generate and configure a group of generated audiences, and use them to receive feedback on their speech during and after the practice in multiple aspects. AudiLens leverages the capability of LLMs in being able to generate a diverse set of personas and being able to simulate human behavior, and provide flexibility to the speaker in terms of practicing their speech with multiple sets of audience groups in multiple speech formats. We demonstrate the use of AudiLens in two scenarios—giving a tutorial and debating.