Available under a Creative Commons Attribution Non-Commercial Share Alike 4.0 International Licence
Convergence of acoustic/prosodic (a/p) features between two speakers is a well-known property of human dialogue. It has been suggested that this particular aspect of human interaction should be implemented in spoken dialogue systems, so that they can be perceived as more “humanlike”. This paper presents a quantitative analysis method that can provide information required for modeling the phenomenon of convergence. The analysis is a combination of TAMA, a previously introduced data extraction method, and bivariate time series analysis. Results show significant correlation of a/p features between speaker dyads in the recorded dialogues analyzed, and indicate a significant,amount of feedback, which a statistical verification of bidirectional convergence.
Kousidis, S. et al. (2009) Convergence in Human Dialogues Time Series Analysis of Acoustic Feature. SPECOM 2009, St. Petersburg, Russian Federation.