The Charismatic Lecturer’s Voice: Explainable Machine Learning Models

This study applies explainable machine learning to identify which vocal attributes in a lecturer’s speech influence students’ views of a lecturer’s charisma, a key contributor to teaching quality. It further explores whether vocal qualities differ between male and female lecturers and how students of different genders respond to these differences, offering insights into voice-related factors that influence the impact of educators. Speech segments from YouTube videos featuring 200 native-English lecturers were evaluated by 900 students using charisma rating scales. A set of attributes related to three primary prosodic dimensions of voice - pitch, rhythm, and loudness - was computed. A random forest classifier was employed to predict the charisma level based on the speech attributes and to list and rank the attributes that contributed most to the prediction. The findings revealed prominent vocal attributes that achieved higher charisma scores in the students' ratings. Same-gender evaluations of charisma were mainly based on pitch, while cross-gender evaluations rely mostly on loudness or rhythm. The automated, interpretable method provides a reliable and efficient way to measure vocal charisma in academic lecturers. It can be adapted to examine additional individual factors that influence the perception of a lecturer’s charismatic presence. It may also be integrated into practice-based tools, designed to support instructors in improving their presentation skills. Our research bridges the fields of applied psychology and computer science to contribute to the development of educational technology.