Dushanthi Madhushika Manamalage
About
I am a PhD researcher at the University of Auckland, where I develop trustworthy AI methods for early depression detection using speech and language. My research combines machine learning, speech processing, multimodal learning, and federated learning to build AI systems that are accurate, interpretable, and privacy-preserving. Ultimately, my goal is to create clinically meaningful AI technologies that support accessible and ethical mental healthcare.
AI for Mental Health
Developing non-invasive screening methods that analyse speech and language patterns related to depression.
Multimodal Learning
Combining audio and text representations to improve robustness, generalisation, and clinical relevance.
Privacy & Interpretability
Studying federated and representation-aware approaches for trustworthy healthcare AI.
Featured Research
Selected research themes from my PhD and related projects.
Multimodal Depression Detection
Speech and text-based modelling for early depression screening.
Privacy-Preserving Learning
Federated multimodal learning for sensitive mental health data.
Privacy-Performance Trade-offs Analysis
Quantifying the trade-off between predictive performance and privacy preservation to support trustworthy federated AI for healthcare.
Recent Highlights
Key publications, media, and academic activities.
Towards Trustworthy Speech-Based Multimodal Depression Detection
Presents my PhD research vision for developing trustworthy AI systems for speech-based depression detection, integrating multimodal learning, privacy-preserving federated learning, and representation-level interpretability.
More of us are using AI are therapists - should we be?
A live media interview discussing the opportunities and limitations of AI in mental healthcare, and how AI can support therapists through early detection, self-reflection, and scalable mental health technologies.
Quality matters: Improving multimodal speech-based depression detection with quality-aware learning
An abstract on using speech as a non-invasive digital biomarker for AI-assisted early depression screening.
AI can’t replace mental health therapists. But here’s where it might make a difference
A public-facing commentary discussing the opportunities and limitations of AI in mental healthcare, and how AI can support therapists through early detection, self-reflection, and scalable mental health technologies.
FedMPA: A Novel Privacy-Performance Optimization Approach for Multimodal Speech-Based Depression Detection
A privacy-preserving multimodal learning framework for speech-based depression detection, designed to balance predictive performance with privacy protection in federated settings.
Let's Connect
I'm always interested in discussing research collaborations, AI for healthcare, and academic opportunities.
