Multimodal Transfer Learning for Oral Presentation Assessment

Oral communication has consistently been ranked as a key skill, with 90 percent of hiring managers and 80 business executives saying it is very important for college graduates to possess, according recent survey. Consequently, training evaluating oral presentation skills remains priority educators worldwide, there are increasing numbers automated tools developed providing feedback assessment such skills. However, modeling approaches typically require collecting large amounts data labels, which can be both expensive laborious. In this paper, we explore the possibility transfer learning between two different but related multimodal datasets benefit evaluation performance. We utilize knowledge from job interview dataset pretraining material adapt learned pre-trained model small amount improve task. demonstrate efficacy our approach, especially in improving performance inference on (< 100 points), report findings. Moreover, give comparison proposed TL approach standard method based large-scale model. Despite simplicity results show that promise application smaller ours.