Auditory_cortex

📂 Repository Structure

This section outlines the main directories and files. See the corresponding documentation for details.

auditory_cortex/
├── auditory_cortex/             ← Core functionality code
│   ├── config/                  ← Configuration files
│   ├── dnn_feature_extractor/   ← DNN feature extraction
│   ├── neural_data/             ← Neural data handling
│   ├── plotters/                ← Plotting utilities
│   ├── analyses/                ← Analysis utilities
│   ├── dataloader.py            ← Data loading
│   ├── encoding.py              ← Encoding models
│   ├── utils.py                 ← Helper functions
├── docs/                        ← Documentation
├── scripts/                     ← Experiment scripts
├── notebooks/                   ← Analysis notebooks
├── Gilbreth_job_submission/     ← Cluster job scripts
└── README.md                    ← Project overview

📚 Documentation

• 🧑‍🔬 User Guide — How to run scripts and use the models
• 👨‍💻 Developer Guide — Contributing, structure, dev notes
• 🎧 Dataset Guide — Audio/neural dataset formats and loading
• 🏋️ Training Guide — Training new models or fine-tuning
• 📈 Evaluation Guide — Metrics, evaluation scripts, visualization

Repo Link

git clone https://github.com/bilalhsp/Auditory_Cortex.git

go to the directory Auditory_Cortex

pip install -e .

git add .
git commit -m "message"
git push origin main

username:
password:

IF memory error

go to the subject directory: rm -rf .*

Modelling the auditory cortex using task optimized Deep Learning models

Dataset.py module defines the Neural_Data class that provides routines for loading the neural data.

Example code to use "Dataset" class:

from auditory_cortex.Dataset import Neural_Data
data = dataset.Neural_Data(path, session)

#"path_of_working_directory" is the address of the folder containing channel files and json file.

Functions:

data.retrieve_spike_times(sent = sentence_code)

Returns times of spikes, relative to stimulus onset or absolute time

provide 'sentence_code' or 'trial_number' as argument

Returns a dictionary, with channel # as the keys to the spike_times of specific channel.

data.retrieve_spikes_count(sent = sentence_code, win=bin_size)

Returns number of spikes in every 'win' miliseconds duration following the stimulus onset time.

provide 'sentence_code' or 'trial_number' as argument

provide 'bin_size' as the desired size of bin in miliseconds

Returns a dictionary, with channel # as the keys to the spikes_count of specific channel.

Installing the virtual environment on Gilbreth

In order to install the virtual environment on Gilbreth..

• Load latest cuda and anaconda module to be able to use conda command on Gilbreth module load cuda/12.1.1
  module load anaconda/2020.11-py38

• Creating virtual environemt as a 'module' for ease of loading on Gilbreth. -p specifies location where packages are installed, -m specifies location where module file is created. conda-env-mod create -p /depot/jgmakin/data/conda_env/cortex_project -m /depot/jgmakin/data/conda_env/etc/modules -j to remove it: conda-env-mod delete -p /depot/jgmakin/data/conda_env/cortex_project -m /depot/jgmakin/data/conda_env/etc/modules -j

• Loading the newly created environment as a module. module purge module load anaconda/2020.11-py38 module use /depot/jgmakin/data/conda_env/etc/modules module load conda-env/cortex_project-py3.8.5

• Install rest of the packages. conda install numpy scipy matplotlib pandas jiwer cupy pytorch=2.0.1 torchaudio=2.0.2 tensorflow=2.13 tensorflow-probability=0.21 -c conda-forge -c pytorch

  pip install tensorflow==2.13 tensorflow-probability==0.21

  module loaded for cudatoolkit=11.8 cudnn=8.6

pretrained 'Wav2Letter':

Link of the website that provides pretrained 'Wav2letter' is: https://github.com/flashlight/wav2letter/tree/wav2letter-lua?tab=readme-ov-file#pre-trained-models

The checkpoint 'wget https://s3.amazonaws.com/wav2letter/models/librispeech-glu-highdropout.bin' is downloaded to: /scratch/gilbreth/ahmedb/wav2letter/pretrained