Discriminating Divergent/Convergent Phases of Meeting Using Non-Verbal Speech Patterns

Abstract

The goal of this paper is to focus on non-verbal speech information during meeting and see if this information contains cues enabling the discrimination of meeting phases—divergent and convergent phases using decision trees. Group task experiments were conducted using a modified 20Q. The recorded speech was analyzed to identify various utterance pattern features—utterance frequency, length of utterance, turn-taking pattern frequency, etc. Discrimination trials were conducted on groups of friends, groups of strangers, and on both groups together using these features, and discrimination accuracy rates were obtained of 77.3%, 85.2% and 77.3%, respectively, in open tests. These results are quite good, considering that they are based on non-verbal speech information alone. Among the features relating to utterance patterns used in this work, we found that silence frequency and quasi-overlapping frequency were especially effective for discrimination. Our results did not find that group friendliness or task difficulty information contributed to effective discrimination of the meeting phases.