View my complete publication list on Google Scholar
The thesis behind MoviSound: three first-author papers on giving teachers feedback from recordings of their own classes, and the platform that let them use it.
View thesisFine-tuned a multilingual DeBERTa model and combined it with features of the speech to classify what the teacher is doing each time they speak, with SHAP to show what drove each decision.
View publicationSystematic review of 82 studies (2014-2024) on using audio processing in educational research.
View publicationClassified how a class is being taught (lecture, group work or Wooclap quizzes) from who speaks when, and tested it on teachers and classrooms the model had not seen.
View publicationBuilt a system to automatically analyze classroom recordings and generate insights for teachers.
View publicationAnalyzed how student response systems work in real classrooms using multimodal AI.
View publication