Developing a Machine Learning Algorithm for Predicting a Speakers Emotional Status and Tendency

Project description

• Knowing the emotional state of another person is generally important, but is vital in situations such as healthcare where people are in close proximity and tensions can be high. Monitoring a patient’s emotional state during a consultation, or attempting to deescalate an evolving situation, is an important skill for healthcare providers. A skill that is not innate in everyone, and so must be taught. It would also be helpful to have such information proceeding an altercation to facilitate the review of case studies. Ultimately, cultivating an empathetic understanding of a patient’s emotional well-being is integral to delivering person-centred care.
• This project aims to develop a machine learning algorithm capable of accurately predicting the emotional status and tendency of speakers based on their speech patterns. An extension goal for this project would be to make the tool bi- or even multi-lingual.

1. Develop a Classification Model: Create a machine learning algorithm that can analyse speech data and predict the speakers’ current emotional status.
2. Develop a Predictive Model: Based on the temporal output of the classification model create an algorithm to predict the likelihood of certain emotional states occurring in the future.
3. Data Collection: Source a diverse dataset of speech recordings annotated with emotional labels. Publicly available databases should be utilised for the training data. Testing data could be publicly available, but will ideally be sourced from clinical settings.
4. Feature Extraction: Investigate pre-processing techniques that will aid in the identification and extraction of relevant features from the speech data that are indicative of emotional states.
5. Model Training and Validation: Train the machine learning model using the pre-processed data and validate its performance using appropriate metrics.
6. Predictive Tendencies: Extend the model to predict not only the current emotional status but also the likelihood of future emotional states based on historical data.
7. Trend Analysis: Analyse trends in emotional states over time to identify patterns and triggers.
8.  Implementation and Testing: Implement the predictive model and test its real-world applicability in simulated scenarios and clinical recordings.
9. Real-time Monitoring (extension goal): make the system run in real-time and provide monitoring capabilities to facilitate immediate feedback and interventions.
10. Muli-lingual Operation (extension goal): collect multi-lingual data and train a classification system for accurate operation in multiple languages.

1. Accurate Predictive Model: A machine learning algorithm that can reliably predict emotional states and tendencies from speech data.
2. Comprehensive Dataset: A well-curated dataset of speech recordings with emotional annotations.
3. Research Publication: A joint research paper detailing the methodology, findings, and implications of the project.
4. Practical Applications: Potential applications of the predictive model in health and social care services, and other relevant fields.

Note: You need to register interest in projects from different supervisors (not a number of projects with the one supervisor).
You must also contact each supervisor directly to discuss both the project details and your suitability to undertake the project.