Menoua Keshishian

This study compares the hierarchical processing of speech in the human auditory cortex with a recurrent ASR model, finding parallel representations across acoustic, phonetic, lexical, and semantic stages.

The paper investigates how human auditory cortex integrates sound structures over time versus structure duration, finding that integration is primarily time-yoked.

This paper introduces a deep neural network framework to model nonlinear sensory responses, showing improved prediction accuracy and revealing distinct nonlinear transformations in auditory regions.