Deriving Music and Speech Response from Deconvolution

This repository is the code for the article "Subcortical responses to music and speech are alike while cortical responses diverge" by Shan et al (2024) (https://doi.org/10.1038/s41598-023-50438-0).

The EEG-BIDS format data for the paper are available on openneuro: https://openneuro.org/datasets/ds004356/versions/2.2.1

Prerequisite Libraries

• mne
• expyfun

Stimuli preprocessing

• stim_preproc.py: music and speech stimulus preprocessing, including converting, trimming the silence, flatten the music envelope.
• spectral_matching_bandpower.py: script for the spectral matching procesing.

Regressor generation

• rectified_regressor_gen.py: script for generating half-wave rectified stimulus waveform as the regressor.
• IHC_ANM_regressor_gen.py: script for generating IHC and ANM regressor. These files must be downloaded before the code can work:
  • Code: 2018 Model: Cochlea+OAE+AN+ABR+EFR (Matlab/Python) from https://www.waves.intec.ugent.be/members/sarah-verhulst (this is for the ic_cn2018 module)
  • UR_EAR_2020b from https://www.urmc.rochester.edu/labs/carney.aspx (this is for importing the cohclea module)
• regressor_click.py: script for generating the aforementioned regressors in response to a click stimulus. The results can be used in the analysis notebook to estimate the subcortical delay based on the cross correlation between the regressors.

Response deriving

• derive_click_ABR.py: deriving click-evoked ABR using cross-correlation method.
• derive_music_speech_ABR: deriving music- and speech-evoked ABR using deconvolution method from three regressors.
• derive_music_speech_cortical: deriving music- and speech-evoked cortical resonses using deconvolution method from three regressors.

Analysis

• SNR.py: calculating the SNR of music- and speech-evoked ABR from the three regressors.
• SNR_recordingtime.py: calculating the averaged ABR derived from different length of recording.
• EEG_predict.py: script for creating predicted EEG data from the kernels from the three regressors using convolution.
• EEG_predict_correlation: calculating the correlation coefficient between predicted EEG and real EEG data.
• spectral_coherence: calculating the spectral coherence between predicted EEG and real EEG data.
• odd_even_morphology.py: calculating the averaged ABR of evenly splitted two set and their waveform correlation (null). This also calculate the correaltion between music- and speech-evoked ABR from ANM regressor, and do the wilcoxon test this with the null distribution.
• analyse_results.ipynb: notebook for loading the generated results and making the correlation and coherence comparison figures.

Example pipeline for subcortical results

The following commands were executed to predict the EEG responses using the different regressors and compute their correlation and coherence:

cd regressor_gen; python rectified_regressor_gen.py; python IHC_ANM_regressor_gen.py; cd ../response_derive; python derive_music_speech_ABR.py; cd ../analysis; python EEG_predict.py; python EEG_predict_correlation.py; python spectral_coherence.py

After running the scripts, the analyse_results.ipynb notebook can be used to plot the results.