Hybrid capstone projects present unique emotional and collaboration challenges for software engineering students working across in-person and remote settings. Emotions significantly influence how students communicate, coordinate, and evaluate their team performance, yet they are often underexplored in software engineering education. This study investigates the emotional dynamics of student teams during Scrum retrospective meetings using speech-based emotion recognition. We analyzed audio from 24 retrospectives across four student teams using WavLM-Large, fine-tuned on the IEMOCAP dataset, to detect expressed emotions. These results were correlated with self-assessed team performance data collected from post-sprint surveys. Our findings reveal that (1) frustration and excitement as the most frequent non-neutral emotions; (2) emotional expression varied significantly across meeting formats, with in-person meetings fostering a wider emotional range due to higher social presence, hybrid meetings displaying mixed emotional dynamics caused by the split between remote and in-person participants, and online meetings being more emotionally restrained due to limited non-verbal cues; and (3) a strong negative correlation between frustration and perceived team performance, while calmer, neutral-toned discussions were positively associated with higher self-assessed ratings. These findings highlight the importance of emotional awareness in team-based learning and offer recommendations for creating emotionally supportive, modality-aware learning environments in software engineering education.

@inproceedings{paperbot11959,
  title = {When Emotions Matter: Speech-Based Emotion Detection in Capstone Project Retrospective Meetings while Working Hybrid},
  author = {Wardah Naeem Awan and Maria Paasivaara and Peter A. Gloor},
  booktitle = {ICSE},
  year = {2026},
  doi = {10.1145/3786580.3786987}
}