AI-based dementia screening using emotion change detection in speech signals

About the Project

With the development of AI tools, including machine/deep learning, speech signal processing has reached a high level of accuracy, both for neutral and expressive speech. Nowadays, it is well established that machines can recognise words and other non-speech human sounds such as laughs, screams and cries. Even though machines can recognise emotions from speech, this is achieved with lesser accuracy than neutral speech.

In case of speech emotion recognition, it is still problematic to detect sudden emotion change, and more particularly link it to some cues in mental disorders. In fact, sudden emotion change during speech is known to be a significant symptom of some mental-health disorders, such as depression and autism. It is also regarded as a precursor sign of some neuro-degenerative disorders such as Alzheimer’s and Parkinson’s diseases, Dementia due to Alzheimer’s and dyslexia.

Therefore, this research aims to answer to the following questions:

1. Firstly, is it possible to model emotion change detection on the fly, i.e. in continuous speech signal stream? This problem has been resolved for other sound event detection, but up to our knowledge, not for emotions, as has been reported in our last survey about Anomalous sound Event Detection.
2. Secondly, can sudden emotion change detection help diagnostic Dementia? If yes, which types of dementia can be identified, and how to model such a correlation using AI models?
3. Thirdly, how to leverage the designed emotion change detection model to develop some AI tools that can provide early detection signs of dementia, and/or help care givers detect imminent deterioration of the mental health state of the attained subject ?

In the light of these research questions, AI in general, and machine/deep learning in particular are potentially reliable to resolve the raised challenges. In fact, speech emotion recognition has always been based on data-driven models, such as Hidden Markov Models (HMM), deep neural networks (DNN), and more recently moving towards end-to-end or on-the-fly models, whereas emotion change/anomaly detection is still awaiting a well-defined machine learning-based framework to deal with the issues mentioned above. Such a framework should be able to model the audio source and meet the goal of the emotion change/anomaly detection process, for application to dementia screening.

The project offers the candidate new opportunities to gain invaluable experience in the relevant areas, including AI, machine learning, audio signal processing and anomaly detection. Collaboration with the Centre for Dementia Studies at the University of Bradford is also projected, where the candidate can work closely with experts in Dementia screening and monitoring.

Eligibility

Candidates are expected to hold (or be about to obtain) a minimum 2:1 honours degree (or equivalent) in a related area / subject, e.g. Computer Science, Data Science, Big Data, AI, Mathematics, etc. MSc, MA or relevant experience in a related discipline is highly desirable.

How to apply

Formal applications can be submitted via the University of Bradford web site. Applicants should register an account, select 'Postgraduate Research' as the course type and use the keywords 'computer science'. Please include the project title on the Research Proposal section; applicants are not required to supply a research proposal for this project.

Informal enquiries are also welcome.