To satisfy the requirements for running experiments please refer to environment.yml file,

conda env create -f environment.yml
conda activate emotionCampaign1

Note: To extract wav2vec features using fairseq, the version used is "1.0.0a0+5273bbb", please check the version, at the time of writing this it is the version on their github and not the one that can be installed by pip. In case of any issue, we also provide our copy of fairseq at the time of use to ensure reproducibility.

Create a new environment on conda and after activating it, run the following commands to get the necessary packages installed:

conda create --name emotionCampaign python=3
conda activate emotionCampaign
conda install pytorch==1.6.0 cudatoolkit=10.1 -c pytorch # Please change cudatoolkit depending on your case (e.g. use "nvidia-smi")!
pip install python_speech_features==0.6
# Follow fairseq installation from: https://github.com/pytorch/fairseq: (in case of issue, we also provide our copy of fairseq at the time of use to ensure reproducibility, it is located near the folder containing the scripts)
# cd fairseq_c
# pip install --editable ./
conda install pandas==1.1.3
conda install -c conda-forge pydub==0.25.1
conda install -c anaconda scipy==1.6.2

Datasets used here were RECOLA and AlloSat, information regarding how to access them can be found below:

RECOLA: https://diuf.unifr.ch/main/diva/recola/download.html

AlloSat: https://lium.univ-lemans.fr/en/allosat/

1- Change directory (cd) to Scripts/DatasetsRunner

2- Open the file and set the annotations, models, features, and the path to those features in the setExperiments function. It is already filled with some examples as it was used for LeBenchmark, so please feel free to change those.

3- Run the file:

python RECOLA_2016.py -w "[path_to_datasets]/RECOLA_2016/recordings_audio" -a "[path_to_datasets]/RECOLA_2016/ratings_gold_standard" -o "[path_to_datasets]/RECOLA_2016"

NOTE: The mentioned file only works with the version of RECOLA provided for AVEC_2016 challenges and it requires the ratings_gold_standard and recordings_audio to be under [path_to_datasets]/RECOLA_2016.

1- Change directory (cd) to Scripts/DatasetsRunner

2- Open the file and set the annotations, models, features, and the path to those features in the setExperiments function. It is already filled with some examples as it was used for LeBenchmark, so please feel free to change those.

3- Run the file:

python AlloSat.py -w "[path_to_datasets]/AlloSat_corpus/audio" -a "[path_to_datasets]/AlloSat_corpus/annotations/labels" -p "[path_to_datasets]/AlloSat_corpus/" -o "[path_to_datasets]/AlloSat"

NOTE: The mentioned file only works with the version of AlloSat_corpus provided as is. However, it can also be used as an example to be used for other datasets.