Code repository

Code repository for the manuscript "Distinct Neural processes link speech planning and execution" by Duraivel et al (2026).

This repository contains all MATLAB scripts needed to reproduce the figures in the manuscript. Data are available on Zenodo (doi: 10.5281/zenodo.20720848).

If using this code, please cite the corresponding paper

Overview

We used intracranial EEG (sEEG/ECoG) to investigate how the human brain plans and executes speech sequences during articulation. Using temporal generalization decoding and high-gamma (70–150 Hz) cortical activation analyses across 10 anatomically defined regions of interest, we identified a two-tiered hierarchical structures in three speech networks:

• Planning network (IFG, rMFG, cMFG)
• Articulation network (SMC, IPC)
• Monitoring network (Insula, aSTG, pSTG, STS, PAC)

---

Repository Structure

speechPlanning_manuscript/
├── setup.m                    # Configure paths — edit before running anything
├── data/
│   ├── colors.mat             # Shared color palette for all plots
│   └── README.md              # Zenodo package description and figure index
├── visualization/             # One script per figure — run these to reproduce figures
│   ├── figure1_behavior.m
│   ├── figure2_hg_timecourses.m
│   ├── figure3_syllable_decoding.m
│   ├── figure4_hierarchy.m
│   ├── figure5_phoneme_sequence.m
│   ├── extended_data1_rh_hg.m
│   ├── extended_data2_syllable_roi_lh.m
│   ├── extended_data3_hierarchy_per_roi.m
│   ├── supp1_behavior_lme.m
│   ├── supp2_hg_go_cue.m
│   ├── supp3_nnmf.m
│   ├── supp4_syllable_rh_roi.m
│   ├── supp5_syllable_go_cue.m
│   ├── supp6_phonotactics.m
│   ├── supp7_ifg_specificity.m
│   ├── supp8_subject_specific.m
│   ├── supp9_micro_ecog.m
│   ├── supp10_phoneme_rh_roi.m
│   └── supp11_phoneme_sequence_rh.m
├── processing/                # Scripts that generate the .mat result files from raw data
│   ├── electrode_selection.m
│   ├── nonNegMatFact.m
│   ├── Phoneme_Sequencing_*.m
│   └── server/                # HPC batch scripts (SLURM/cluster jobs)
└── decoding/                  # µECoG-specific decoding (requires raw recordings)
    ├── setup_micro.m
    ├── figure5_micro_s14/
    ├── supp9_micro_s16_s62/
    └── utils/

---

Requirements

• MATLAB R2020a or later
• IEEG_Pipelines —

---

Setup

1. Clone this repository

git clone https://github.com/coganlab/speech-planning.git
cd speech-planning

2. Download the data package from Zenodo

Download the Zenodo archive (doi: 10.5281/zenodo.20720848) and extract it to a local folder. The package contains four subdirectories:

speech_planning_data/
├── hg_activations/      # Pooled high-gamma epoch data
├── decoding_results/    # Temporal generalization decoder outputs
├── behavior/            # Behavioral summary files
└── micro_ecog/          # µECoG electrode metadata

See data/README.md for a full description of every file in the package.

3. Configure paths in setup.m

Open setup.m and set the two user-configurable paths at the top:

DATA_DIR = '/path/to/speech_planning_data';   % folder extracted from Zenodo
IEEG_DIR = '/path/to/IEEG_Pipelines/MATLAB';  % your local clone of IEEG_Pipelines

HG_DIR, DEC_DIR, BEH_DIR, and OUT_DIR are derived automatically from DATA_DIR.

4. Run a figure script

In MATLAB, navigate to the repository root and run any visualization script:

cd /path/to/speech-planning
visualization/figure3_syllable_decoding   % no iEEG access needed

Figures are saved as .eps and .png to OUT_DIR/figure3/ (i.e., inside your Zenodo data folder under figures/).

---

Reproducing Each Figure

Figure	Script	Zenodo data needed	Requires iEEG access?
Figure 1	visualization/figure1_behavior.m	—	Yes (popTaskSubjectData)
Figure 2	visualization/figure2_hg_timecourses.m	hg_activations/	Yes (extractRoiIds)
Figure 3	visualization/figure3_syllable_decoding.m	decoding_results/	No
Figure 4	visualization/figure4_hierarchy.m	decoding_results/	No
Figure 5	visualization/figure5_phoneme_sequence.m	decoding_results/	No (µECoG panels: Yes)
Extended Data 1	visualization/extended_data1_rh_hg.m	hg_activations/	Yes
Extended Data 2	visualization/extended_data2_syllable_roi_lh.m	decoding_results/	No
Extended Data 3	visualization/extended_data3_hierarchy_per_roi.m	decoding_results/	No
Supp 1	visualization/supp1_behavior_lme.m	—	Yes
Supp 2	visualization/supp2_hg_go_cue.m	hg_activations/	Yes
Supp 3	visualization/supp3_nnmf.m	hg_activations/	Yes
Supp 4	visualization/supp4_syllable_rh_roi.m	decoding_results/	No
Supp 5	visualization/supp5_syllable_go_cue.m	decoding_results/	No
Supp 6	visualization/supp6_phonotactics.m	decoding_results/	No
Supp 7	visualization/supp7_ifg_specificity.m	hg_activations/	Yes
Supp 8	visualization/supp8_subject_specific.m	hg_activations/	Yes
Supp 9	visualization/supp9_micro_ecog.m	micro_ecog/ + raw µECoG	Yes
Supp 10	visualization/supp10_phoneme_rh_roi.m	decoding_results/	No
Supp 11	visualization/supp11_phoneme_sequence_rh.m	decoding_results/	No

Figures requiring institutional iEEG access use popTaskSubjectData from IEEG_Pipelines to load raw trial data from the CoganLab data server. These figures cannot be reproduced from the Zenodo package alone.

---

µECoG Scripts

Scripts in decoding/ use raw µECoG recordings and require additional setup:

1. Open decoding/setup_micro.m
2. Set MICRO_DIR to your local Micro/Processed Data folder
3. Set RECON_DIR to your local clone of ECoG_Recon (matlab_code subfolder)

µECoG subjects:

• S14 (manuscript S3, 256-ch) — Figure 4 and 5 µECoG panels → decoding/figure5_micro_s14/
• S16 (manuscript S1, 128-ch) — Figure 3 µECoG panels → decoding/supp9_micro_s16_s62/
• S62 (manuscript S2, 256-ch) — Figure 3 µECoG panels → decoding/supp9_micro_s16_s62/

Electrode metadata for these subjects (impedance, significant channel masks) is included in the Zenodo package under micro_ecog/.

---

Regenerating Result Files from Raw Data

The processing/ directory contains scripts that generate the .mat files in the Zenodo package from raw iEEG recordings. These require institutional data access and are provided for transparency, not for general reproduction.

• electrode_selection.m — identifies production-active electrodes (FDR q<0.05)
• Phoneme_Sequencing_timeGen_syllable_ROI_pooledSubjects.m — runs syllable temporal generalization decoder per ROI
• Phoneme_Sequencing_timeGen_Blick_ROI_pooledSubjects.m — runs phonotactic (BLICK) regressor
• nonNegMatFact.m + Phoneme_Sequencing_timeGen_decomposition_pooledSubjects.m — NNMF decomposition (k=5)
• processing/server/ — parallelized HPC versions of the above for cluster submission

---

Data Availability

Processed neural data are available on Zenodo under restricted access: doi: 10.5281/zenodo.20720848. Access can be requested via the Zenodo record page.

Raw iEEG recordings are available upon request to corresponding authors.

---

Contact

Suseendrakumar Duraivel | CoganLab, Duke University

Gregory Cogan | CoganLab, Duke University