Acoustic and linguistic ecoding analysis pipelines for mTRF models based on natural continuous speech

This repository hosts the data processing pielines and statistical models described in our paper entitled Neural encoding of linguistic speech cues is unaffected by cognitive decline, but decreases with increasing hearing impairment, doi:10.1038/s41598-024-69602-1.

We share the code for reasons of transparency. The data—Electroencephalography (EEG) and audio files—cannot be shared, but are available upon request. The code is adapted to our environment and data infrastructure and cannot be executed without adjustments.

Structure

• preprocessing/: This directory contains the EEG and speech preprocessing pipelines, organized into the following subdirectories:

  • preprocessing/eeg/: Contains the pipelines used for preprocessing the EEG data for multivariate Temporal Response Function (mTRF) analysis.
  • preprocessing/speech/: Comprises two subdirectories:
    • preprocessing/speech/representations/: Includes pipelines for computing linguistic markers (segmentation, word-based, and phoneme-based speech features for the mTRF models) from the output of the forced aligner. A different environment is required to run these scripts, detailed in linfeatures.yml.
    • preprocessing/speech/features/: Contains pipelines to generate acoustic and linguistic time series as speech features for the mTRF model.

• boosting/: This directory holds all scripts used to set up the mTRF models using techniques from the Eelbrain Toolbox. For a detailed explanation and methodology, refer to Brodbeck et al. (2023).

• statistics/: This folder contains the R and Python scripts used to run the statistical models and perform the descriptive statistics reported in the main manuscript and supplementary material. These scripts ensure reproducibility and transparency of our data analysis processes.

Environment setup

The environment.yml file included in this repository contains the specifications for the conda environment used in this project. Please note that the code is adapted to our environment and data infrastructure and cannot be executed without adjustments.