Speech Analytics on Individual and Group Audio Data to Understand Collaboration

Abstract

Collaborative learning in classrooms require instructors to monitor student groups to ensure they make progress with the tasks. One way learning analytics has helped facilitating such classrooms is by providing speech-based solutions to help instructors monitor. In this poster, we investigate two different ways of collecting audio data from group work namely, group audio data and individual audio data and how voice activity detection (VAD) can be used to predict student collaboration. Both types of audio data were collected from classes focused on collaborative problem solving that were part of an introductory undergraduate engineering course. Preliminary analysis of 8 groups of audio data using VAD indicate that individual audio data could provide information regarding turn ending, turn overlap, and turn duration of individual students which can be critical in understanding the quality of collaboration of a group which cannot be obtained consistently using group audio data.