[detection/asvspoof5/v11_ssl_gmlp] add folder

[zlin0]

This is a draft PR. The performance on ASVspoof 5 is not good and requires further exploration of suitable parameters.

[Copilot]

Pull Request Overview

This PR adds a new experimental configuration folder for ASVspoof5 detection using SSL (Self-Supervised Learning) with gMLP (gated Multi-Layer Perceptron) models. The implementation explores different SSL model variants (WavLM, wav2vec2, XLS-R) with frozen and fine-tuned configurations for audio spoofing detection.

• Adds complete training pipeline with data preparation, model training, evaluation, and performance measurement stages
• Implements 6 different SSL model configurations combining various pre-trained models with frozen/fine-tuned settings
• Provides SLURM-based distributed training setup for HPC environments

Reviewed Changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
wedefense	Symbolic link to wedefense framework
tools	Symbolic link to shared tools directory
slurm_train.sh	SLURM job script for distributed training with model configuration array
run.sh	Main training pipeline script with 7 stages from data prep to evaluation
path.sh	Environment setup script for Python paths
local	Symbolic link to shared local utilities
conf/*.yaml	6 SSL model configurations for different wav2vec2/WavLM/XLS-R variants

[Copilot]

The symbolic link target '/eval' is an absolute path which will create a link in the root directory instead of the intended relative path. This should be 'eval' without the leading slash.

Suggested change

	ln -s flac_E_eval /eval
	ln -s flac_E_eval eval

[Copilot]

Using sed to replace spaces with tabs is fragile and could break if filenames contain spaces. Consider using awk or a more robust text processing approach to format the cm_key_file.txt properly.

Suggested change

	sed -i "s/ /\t/g" ${data}/${dset}/cm_key_file.txt
	awk '{$1=$1; gsub(" ", "\t"); print}' ${data}/${dset}/cm_key_file.txt > ${data}/${dset}/cm_key_file.tmp && mv ${data}/${dset}/cm_key_file.tmp ${data}/${dset}/cm_key_file.txt

[Copilot]

The email address is hardcoded, which exposes personal information and makes the script non-portable. Consider parameterizing this or removing it from the committed code.

Suggested change

	#SBATCH --mail-user="lzhan268@jh.edu" #email for reporting
	#SBATCH --mail-type=END,FAIL #report types
	#SBATCH --output=./logs/slurm-%A_%a.out
	#SBATCH --array=4-4
	#SBATCH --mail-user="${SLURM_MAIL_USER}" #email for reporting
	#SBATCH --mail-type=END,FAIL #report types
	#SBATCH --output=./logs/slurm-%A_%a.out
	#SBATCH --array=4-4
	if [ -z "${SLURM_MAIL_USER}" ]; then
	echo "Error: SLURM_MAIL_USER environment variable is not set. Please set it to your email address."
	exit 1
	fi

[Copilot]

The hardcoded absolute path contains a username 'lzhan268' making it non-portable across different users and systems. This should be parameterized or use a relative path.

Suggested change

	ASVspoof5_dir=/export/fs05/lzhan268/workspace/PUBLIC/PartialSpoof/database
	: "${ASVspoof5_dir:?Environment variable ASVspoof5_dir must be set to the ASVspoof5 database directory.}"