Application of Acoustic Feature Recognition Technology for NPC Emotion Expression in Games

Abstract

Speech recognition games are progressing day by day, mainly through the improvement of natural language processing technology, so that even ambiguous utterances of the player can be processed. However, existing game productions have not taken into account what is described as “tone of voice” which can contain information about the speaker's intentions and emotions. Therefore, this study focuses on the human tone of voice. We developed SEReN, a system that infers emotions from players' speech and changes NPCs' reactions and responses according to estimated emotions. Subsequently, we conducted system evaluation experiments. The emotional connection to the NPC character was evaluated using a questionnaire. The study revealed a significant difference between SEReN and the conventional system for some of the questions.