Recent research insights from the TRANSMIXR Academics
In the rapidly evolving fields of virtual reality (VR) and multimedia, understanding and enhancing the Quality of Experience (QoE) for users is of paramount importance. Two recent scientific publications from the TRANSMIXR team at TUS were presented during the annual QoMEX 2024 Conference in Karlsham, Sweden. Both studies explored the impact of emotional dynamics in VR and the predictive power of physiological signals in multimedia experiences.
Unveiling Emotional Dynamics in Collaborative VR
The paper titled “Voices Unveiled: Quality of Experience in Collaborative VR via AssemblyAI and NASA-TLX Analysis,” authored by Bhagyabati Moharana, Dr. Conor Keighrey, and Dr. Niall Murray, investigates the critical role of communication and workload on the quality of experience (QoE ) in collaborative VR tasks (Fig. 1). This work leveraged AssemblyAI for voice sentiment analysis and NASA-TLX for assessing workload. The research involved real-time capturing of conversations and self-reported frustration levels during VR puzzle tasks.
Key findings from the study reveal significant correlations between voice sentiments, task performance, and user frustration. Positive voice sentiments were found to correlate with faster task completion times, while negative sentiments were associated with higher frustration levels. These insights highlight the importance of emotional dynamics in VR environments, suggesting that integrating sentiment analysis into VR systems can significantly enhance user experience and performance. The details of this study can be accessed here.
Predicting quality of experience (QoE) with Physiological Signals in Multimedia Experiences
Another significant contribution to the field comes from the paper “The Role of ECG and Respiration in Predicting Quality of Experience,” authored by Sowmya Vijayakumar under the supervision of Dr. Niall Murray. This research explores the use of Electrocardiogram (ECG) and respiration (RSP) signals to predict QoE, employing various machine learning techniques (Fig. 2).
The study focuses on the effects of hyperparameter tuning, dimensionality reduction, and feature selection algorithms on classification performance. The findings reveal that the Random Forest (RF) classifier was the most effective model for classifying overall and audio quality. Notably, combining ECG and RSP data with an RF classifier achieved the highest F1-score of 87.91% for perceived audio quality. Furthermore, the RF model using only ECG data attained the highest classification F1-score at 80.49% for perceived overall quality, highlighting the importance of ECG signal morphological features.
This research underscores the potential of physiological signals, particularly ECG and RSP, in predicting QoE in multimedia experiences. By demonstrating the effectiveness of these signals in QoE prediction, the study opens avenues for further research to refine and enhance QoE prediction methodologies, ultimately aiming to create more immersive and satisfying user experiences. The details of this study can be accessed here.
Authors: TUS – The Technological University of the Shannon
