Research Study

We are an interdisciplinary research group at the University of Strathclyde working at the intersection of AI and inclusive education. Our goal is to lead the development of ethical, human-centred AI technologies that promote equity, accessibility, and diversity in education and society. Through collaboration across science, engineering, humanities, and social sciences, we design real-world AI solutions that empower learners with additional support needs, reduce inequalities, and inform inclusive educational practices and policy. Take a look at our ongoing research studies below.

AI-Based Emotional Assessment For Students With Additional Support Needs

The research project, “AI-based Emotion Assessment for Students with Additional Support Needs”, under the guidance of Dr Andrew Abel and Dr Laibing Jia, is driven by a commitment to understanding and supporting the emotional well-being of pupils with additional support needs within Scottish classrooms. Human emotion is a complex, multimodal phenomenon, expressed not only through overt behaviours and facial expressions but also via physiological signals such as heart rate. For students who require additional support, particularly those who are neurodivergent, traditional observation techniques often fail to capture the full spectrum of emotional responses. Pupils with autism spectrum disorder, for example, may communicate emotion in ways that diverge significantly from neurotypical patterns, while those with ADHD or dyslexia might present subtle cues that are easily missed by educators. Recent advances in artificial intelligence have enabled the development of multimodal AI models capable of integrating diverse signals, ranging from facial micro-expressions to biometric data, thus offering the potential for a more holistic approach to emotional assessment. However, achieving high levels of accuracy alone is insufficient; the real challenge is to ensure that these models are attuned to the individual and cultural contexts in which they are deployed so that they may be both effective and genuinely accepted. Therefore, this work aims to foster classroom environments where all pupils, regardless of how they express their emotions, feel seen, understood, and supported.

AI in Inclusive Education: Teacher’s Perspective

The study aims to investigate and enhance the optimisation of Artificial Intelligence (AI) tools to support teacher effectiveness and foster equitable educational environments within inclusive classrooms in Scotland. By exploring teachers’ perceptions, identifying key challenges and professional development needs, and examining the design and implementation strategies of AI tools, this research seeks to contribute comprehensive insights into how AI can be harnessed to improve student experiences, achievements, and participation. The ultimate objective is to inform both educational policy and practical applications, ensuring that AI integration not only supports educators but also promotes fairness and inclusivity for all students, thereby addressing and mitigating existing biases and inequalities within the educational landscape.

Classroom Engagement Analysis and Feedback Using Multimodal Emotion Recognition

This project focuses on identifying student engagement in classrooms, a vital factor that influences both learning outcomes and teaching effectiveness. High levels of engagement are strongly linked to improved academic performance. In contrast, low engagement can lead to challenges such as reduced achievement, higher dropout rates, student alienation, and a general lack of interest or motivation. Despite its importance, accurately measuring engagement in classroom environments, especially those with a large number of students, remains a significant challenge. Traditional approaches, such as teacher observations, surveys, and student feedback, are often time-consuming, subjective, and susceptible to bias. From an educator’s perspective, gaining a clear understanding of classroom engagement levels is critical for refining teaching methods and enhancing learning materials.

This research explores the application of deep learning techniques to identify classroom engagement. By utilising cameras to unobtrusively record classroom sessions with the full consent of all participants, we analyse facial expressions, body movements, and other visual signals in a multimodal framework to assess levels of classroom engagement . This approach is especially valuable in large classroom environments, where it is often challenging for educators to monitor each student’s engagement in real-time. The insights generated through this system can support teachers in understanding individual and collective student needs more effectively. In turn, educators can make informed adjustments to their teaching strategies and methods, ultimately fostering a more dynamic, engaging, and responsive learning experience for all students. Our goal is to enhance the overall learning experience by making classrooms technology-enabled, empowering teachers with better tools while upholding the trust and rights of all individuals involved.

Analysis of Automatic Speech Recognition Systems Response to the Lombard Effect, and Increasing System Robustness to Noisy Environments

This study is concerned with evaluating automatic speech recognition (ASR) deep learning models for their ability to handle noise present during speech. In particular, the Lombard Effect is focused on, a phenomenon where people change aspects of their speech such as their volume and mouth movements in the presence of a noisy environment. The Lombard effect is under-explored in current ASR research, and handling noise is a broader issue in the field of speech recognition.

By analysing how varying ASR models perform when handling noisy and Lombard altered audio, it is hoped that the model architecture that increases robustness can be identified. The goal of this research is the creation of an ASR model specifically made to handle noisy speech. Multi-modal support, like video, may also be included for speech too noisy to be processed by audio alone. This robust model would be paired with a clean audio model to create a system for transcription that handles both clean and noisy speech efficiently, utilizing a router to pick which model to use for a given input. This would hopefully create a smaller, more robust system with parts that come together to form a greater whole.