Temporal Response Function-Driven Representational Similarity Analysis for Speech Perception Decoding with MEG and EEG

Abstract

Title: Simple Summary Understanding how the human brain processes and comprehends various sounds remains a major challenge in modern science. Traditional multivariate analysis methods often employ static strategies and overlook the temporal dependencies of neural responses. Therefore, our primary objective was to integrate system identification principles into multivariate pattern analysis. By modeling the relationship between stimulus features and neural responses, we developed a dynamic representation analysis method that successfully separated meaningful sound-driven brain activity from background neural noise. The results demonstrated that this approach offered much higher sensitivity in decoding different categories of sounds, revealing that the human brain showed a distinct preference for biological sounds. Furthermore, it uncovered a much wider auditory network than previously thought, involving not just traditional auditory areas but also limbic and deep brain structures. This work deepens our understanding of the spatiotemporal dynamics of sound processing and highlights the potential of this method for objective clinical assessment and diagnosis.