Master of Science Pegah Salehi will Wednesday October 1st, 2025, at 12:15 hold her Thesis Defense for the PhD degree in Science. The title of the thesis is:
« Visual Realism in AI-Driven Virtual Training Environments for Child Investigative Interviews »
The imperative to safeguard children from abuse demands investigators who can conduct sensitive, evidence-gathering interviews with skill and empathy. Yet conventional lectures, role-plays and actor-based simulations struggle to provide the interactive, visually convincing practice necessary for mastery. This thesis therefore asks: How can artificial intelligence be used to build a visually realistic child-avatar system for investigative-interview training?
Using a Design Science Research Methodology, I iteratively designed, implemented and evaluated a multimodal AI platform that links natural-language processing, speech-to-text, text-to-speech and advanced avatar-generation techniques. Five research strands structure the work: (1) charting technical and ethical challenges in creating child avatars; (2) testing how different levels of visual realism shape learning and acceptance; (3) comparing 2-D desktop and 3-D virtual-reality arenas; (4) optimising talking-head models through faster audio-feature extraction; and (5) endowing avatars with real-time emotional expressiveness.
Empirical studies with child-protection and police professionals show that photorealistic graphics heighten perceived authenticity, whereas stylised characters can ease the “uncanny valley” in dialogue. Three-dimensional VR markedly boosts immersion and emotional resonance, while 2-D interfaces remain more comfortable for extended sessions. Improved audio pipelines cut latency and align lip-sync, and expressive faces help trainees recognise subtle psychosocial cues. Together these contributions deliver a scalable framework for more accessible, engaging and effective training, ultimately aiming to strengthen the quality of child testimony and reduce interview-related trauma.
Supervisory Committee:
1st Opponent: Professor Cheng-Hsin Hsu, Department of Computer Science, National Tsing Hua University, Taiwan
2nd Opponent: Professor Letizia Jaccheri, NTNU
Internal member and leader of the committee: Professor Gunnar Hartvigsen, IFI, UiT
The defence and trial lecture will be streamed from these following links at Panopto:
Defence (12:15 - 16:00)
Trial Lecture (10:15 - 11:15)
The thesis is available Here