Estimation of Presentations Skills Based on Slides and Audio Features

Abstract

This paper proposes a simple estimation of the quality of student oral presentations. It is based on the study and analysis of features extracted from the audio and digital slides of 448 presentations. The main goal of this work is to automatically predict the values assigned by professors to different criteria in a presentation evaluation rubric. Machine Learning methods were used to create several models that classify students in two clusters: high and low performers. The models created from slide features were accurate up to 65%. The most relevant features for the slide-base models were: number of words, images, and tables, and the maximum font size. The audio-based models reached up to 69% of accuracy, with pitch and filled pauses related features being the most significant. The relatively high degrees of accuracy obtained with these very simple features encourage the development of automatic estimation tools for improving presentation skills.

An Active Data Representation of Videos for Automatic Scoring of Oral Presentation Delivery Skills and Feedback Generation

“…Luzardo et al performed two-class (good or poor) classification experiments to predict quality of slides (SQ), SCE, RIGP, and AVV. For predicting SCE, AVV, and RIGP, Luzardo et al used the audio features (minimum, maximum, average and standard deviation of pitch calculated for each student presentation/video) for the two-class (good or poor) classification task, which resulted in an accuracy of 63% (SCE), 69% (AVV), and 67% (RIGP) (Luzardo et al, 2014). Chen et al proposed a different approach.…”

“…The aforementioned studies (Chen et al, 2014;Luzardo et al, 2014;Ochoa et al, 2014;Haider et al, 2016a) analyzed statistics (mean, median values, etc.) of acoustic features over a presentation and showed that acoustic features can predict some of the presentation delivery skills.…”

“…In this new approach, if the focus is, for instance, to provide feedback about gestures, the system would identify those clusters which include gestures related to positive/negative performance through statistical testing between clusters and video-labels. Previous studies conducted on the MLA dataset (Chen et al, 2014;Luzardo et al, 2014;Ochoa et al, 2014) used a statistical representation of audiovisual features (e.g., mean and standard deviation values of fundamental frequency for full presentation) to create a data representation of a full presentation video for a prediction task. Our contention is that the human social behavior varies over time in a presentation and modeling that behavior by averaging low-level features for a full presentation using statistical functions (e.g., mean and standard deviation values of pitch for a full presentation) will result in poor modeling of social signals.…”

Mirroring Teachers' Assessment of Novice Students' Presentations through an Intelligent Tutor System

“…This is the case of [14], who matched the extracted features of slide presentation's with human evaluation, using a classifier. The classifier used features that can be automatically extracted such as: font size, number of words, images, charts and the image entropy of each slide for measuring the contrast.…”

Designing An Automated Assessment of Public Speaking Skills Using Multimodal Cues

“…A summary of these studies can be found in Ochoa, Worsley, Chiluiza, and Luz (2014). In particular, Luzardo, Guamn, Chiluiza, Castells, and Ochoa (2014) investigated using audio cues -i.e., the speaking rate, as estimated from audio files, and some basic prosodic analysis -to judge delivery performance. Echeverría, Avendaño, Chiluiza, Vásquez, and Ochoa (2014) investigated using Kinect motion traces for me asuri ng body language performance during presentations.…”