Perception of Response Time in Dialogue Generation AI with Various Fillers

Abstract

Interactive character content using sentence generation AI recognizes the user's utterance and then generates a response, resulting in unnatural pauses. As a countermeasure, there is a method of inserting a hammer or filler immediately after the end of the user's speech, but the general-purpose hammer is used extensively, which conversely impairs the humanity of the character. Therefore, we proposed a method to make the filler response conform to the content of the user's utterance, and as a result of experiments, it was found that it had a positive impact on the interaction experience.