Exploring AI-assisted methods and acoustic analysis in finding epistemic emotions in learning discussions

When studying learning, transcribing verbal and nonverbal information can be highly time-consuming and laborious. AI methods can speed up this process but reviewing data remains essential. This study examines two approaches that could make learning research less laborious to the researcher. First, it was explored whether acoustic analysis is suitable for picking out possible moments of learning from audio recordings, and second, whether possible moments of learning can be picked out from AI transcribed data of audio recordings. Two pre-service physics teachers’ audio recordings in classroom peer learning sessions were used as source material. The audio was transcribed by a human and an AI tool. From the transcripts, speech turns indicating epistemic emotions and content were picked. With acoustic analysis of the audio, speech turns with high pitch and/or intensity were marked. The results showed that the AI tool experienced multiple challenges in identifying speakers and speech correctly, leaving a vast need for ensuring the content of the speech turns. However, acoustic analysis picked the speech turns indicating epistemic emotions and content with considerable likelihood. This study suggests that acoustic analysis is a very promising method for locating possible learning moments for further study even from real-world audio data.