|Host Organisation||Sorbonne University|
The objective of the internship is to develop an interaction model for a group of social agents. This objective is divided into two distinct parts:
- Analyse models of interpersonal interaction emerging in multi-party interaction of humans.
- Propose new measures to calculate interpersonal synchronization in a group of human agents.
During the 6-month internship, steps 1 and 2 were completed. First of all, by discovering the various works already carried out on interpersonal interactions in a group of agents, both in computer-related work concerning more than specifically the analysis and evaluation of these interactions only in the case of psychological research into the explanation of these behaviours. Then, with this understanding how a group works, how it is organized and the analysis of the different existing methods for measuring interpersonal synchrony, one can propose a brand-new measure of group synchrony from different voice parameters.
Results & Outcomes
Synchrony in multi-party can be computed through four parameters: latency rate, pause duration, speaking time and silence, each adding and improving the accuracy of the calculation.
In order to obtain a calculation of group interpersonal synchrony, it was necessary to first learn the different definitions of the terms of the subject and know the different work already done on it. Thus, the reading of various articles and books on the group psychology, on the different computer or psychological methods for calculating the synchrony, was the beginning of the work.
To work on interpersonal synchrony and get results for to test it afterwards, a video corpus was needed. The choice was to take one already belonging to the team: the ‘council of coaches’ corpus. Then, we had to search for the different parameters to calculate the synchrony, we took some audio parameters to start with, the choice was made on the pause time, speaking time, silence, latency and backchannels. A backchannel is a short utterance produced by one person in the conversation while another person in the conversation is talking, in order to show interest. In order to calculate these different parameters, it was necessary to use a tool to extract prosody settings from the corpus videos, allowing to calculate the data mentioned above. For this we used the open source software OpenSmile.
Based on the different parameters calculated, it was necessary to mix the various parameters to obtain a value for interpersonal synchrony. Interpersonal synchrony is an important value in determining whether a group can communicate fluidly and dynamically. The synchrony described here is thus divided into four parts: latency rate, break time, speaking time and silence, each adding and improving the accuracy of the calculation. The set of tests shows that the synchrony obtained is independent of the number of people and videos used, and that the calculation is in sync.