Semi-automatic Analysis of Spoken Interaction Dynamics in Collaborative Design Sessions

The paper proposes an original approach for design protocol analysis that reduces the time and the effort required to extract relevant insights about the interaction dynamics in co-design sessions. Audio recordings are processed with a newly tailored algorithm that recognizes speakers during collaborative design and extracts indexes that describe the behavior of the designers, such as their degree of involvement in the session, their role (e.g. leadership), the length of their verbal interventions as well as the frequency with which they take the floor. The robustness of the approach is estimated by means of the application of the algorithm to three co-design sessions that are different from each other in terms of number and profile of the participants, language spoken and design task. The algorithm proved to be effective: the results of the analysis of spoken interactions from audio recordings are comparable to those obtained with quantitative techniques (precision up to 80% and faster processing time up to 100x).