Recreating Neural Activity During Speech Production with Language and Speech Model Embeddings

TL;DR

This study demonstrates that embeddings from large-scale language and speech models can accurately reconstruct neural activity during speech production, revealing insights into cortical processing.

Contribution

It introduces a novel approach using pre-trained language and speech model embeddings to reconstruct neural signals during speech production.

Findings

High reconstruction accuracy with correlation coefficients 0.79 to 0.99

Embeddings effectively capture spatio-temporal neural dynamics

Method applicable across all study participants

Abstract

Understanding how neural activity encodes speech and language production is a fundamental challenge in neuroscience and artificial intelligence. This study investigates whether embeddings from large-scale, self-supervised language and speech models can effectively reconstruct high-gamma neural activity characteristics, key indicators of cortical processing, recorded during speech production. We leverage pre-trained embeddings from deep learning models trained on linguistic and acoustic data to represent high-level speech features and map them onto these high-gamma signals. We analyze the extent to which these embeddings preserve the spatio-temporal dynamics of brain activity. Reconstructed neural signals are evaluated against high-gamma ground-truth activity using correlation metrics and signal reconstruction quality assessments. The results indicate that high-gamma activity can be effectively reconstructed using large language and speech model embeddings in all study participants, generating Pearson's correlation coefficients ranging from 0.79 to 0.99.