Oral Presentation: Investigative Interview Training using an AI-Driven Talking Avatar

Title: Investigative Interview Training using a Multimodal Child Avatar

Abstract:

Effective child investigative interview training is needed to bridge the gap between best-practice-guidelines and practice in field. We aim to develop an empirically informed training system using dynamic multimodal child-avatars based on game engine and artificial intelligence (AI) methods including dialogue models trained on well-designed mock interviews with children aged 5-7. Multimodal avatars will enable interviewers to develop and maintain complex skills needed. Combining the dynamic avatar with feedback may be the future of training with availability and cost-efficacy.

We will present “Ilma”, the avatar, that recognizes interviewers’ verbal questioning, automatically responding to prompts in turn-taking manner. Two user studies have been conducted. Prototype (i) with animated-based avatar using virtual reality (VR) was tested by CPS-experts and students finding support for further development. After investigating user’s perception of realism in multiple virtual child-avatars, we will test prototype (ii) with best rated animated-based avatar and synthetic best-rated voice in April 2022, running interviews with CPS-experts and students (n = 50) on text-only, audio-only, desktop monitor VR (2D) and VR (3D) in randomized order. Scenario description is given pre-training, and post-assessment of user-experience (e.g., presence, dynamics, ease of use) and learning effects will be measured. Results from the study will be presented.