Abstract
The proliferation of remote meetings after the pandemic and lockdown has significantly increased the importance of voice-based communication. In situations where visual information is limited, it becomes difficult to interpret emotions of others, and voice-based recognition and analysis of emotions based on voice are extremely important for enhancing the quality of communication. In this study, we aimed to share new insights on remote communication by analyzing dialogue data using a speech emotion recognition system, which we have developed using Valence-Arousal-Dominance Model proposed by Mehrabian and Russell. We utilized the Utsunomiya University Spoken Dialogue Database for Paralinguistic Information Studies to compare human evaluations of emotions with the estimations made by the speech emotion recognition system. To deeply understand the emotional interactions between interlocutors, we conducted a cross-correlation analysis and investigated the time lags in emotions between speakers. Through this analysis, we captured the dynamics of emotions between the interlocutors and revealed that the mutual influence of emotions gradually weakens over time.
