Speaker Differentiation in AAC Data Logging Using Deep Learning

Abstract

High-technology augmentative and alternative communication (AAC) devices are essential tools for individuals with complex communication needs. Automated data logging in these devices enables researchers and clinicians to analyze user performance. However, existing systems cannot distinguish between users when multiple individuals access the same device, compromising the validity of data logs and complicating performance evaluation.

This paper proposes a deep neural network-based visual analysis approach to address this limitation. By processing video recordings of practice sessions, the method detects and identifies different AAC users, ensuring that data logs accurately reflect individual contributions. This solution has the potential to significantly improve the validity of performance data, streamline analysis, and ultimately enhance AAC outcome measures.

Through a combination of advanced video processing and neural network techniques, this approach represents a major step forward in AAC research and clinical practice. It addresses a critical gap in current data logging systems and paves the way for more accurate, user-specific performance evaluation.