ASVspoof

Timelines

[20/10 - 26/10]

[1] Shentong/Haofan: python reimplementation of CQCC

[2] Chi: X-vector feature extraction

[3] Pinxu: GMM/SVM classifier build

Reports

proposal overleaf: https://www.overleaf.com/project/5f6954ebb68ba500013c78f1

Dataset (LA)

Google drive: https://drive.google.com/file/d/1UGs1o2mDiBO9_iaN-0FupS8x0Tkb4xmt/view?usp=sharing

Feature Extraction

CQCC (baseline)

python3 data_processing.py --data_path ./LA/ASVspoof2019_LA_train/flac --output_path ./data/train_cqcc.pkl --label_path ./LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.train.trn.txt --feature_type cqcc

python3 data_processing.py --data_path ./LA/ASVspoof2019_LA_dev/flac --output_path ./data/dev_cqcc.pkl --label_path ./LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.dev.trl.txt --feature_type cqcc

python3 data_processing.py --data_path ./LA/ASVspoof2019_LA_eval/flac --output_path ./data/eval_cqcc.pkl --label_path ./LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.eval.trl.txt --feature_type cqcc

The saved pickle file has the format: [(cqcc_vec[timestepx60], label[bonafide/spoof]) x N instances]

MFCC

python3 data_processing.py --data_path ./LA/ASVspoof2019_LA_train/flac --output_path ./data/train_mfcc.pkl --label_path ./LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.train.trn.txt --feature_type mfcc

python3 data_processing.py --data_path ./LA/ASVspoof2019_LA_dev/flac --output_path ./data/dev_mfcc.pkl --label_path ./LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.dev.trl.txt --feature_type mfcc

python3 data_processing.py --data_path ./LA/ASVspoof2019_LA_eval/flac --output_path ./data/eval_mfcc.pkl --label_path ./LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.eval.trl.txt --feature_type mfcc

The saved pickle file has the format: [(mfcc_vec[timestepx13], label[bonafide/spoof]) x N instances]

Classifier

GMM

SVM

python3 SVM.py --data_path ./data/train.pkl --feature_type mfcc

python3 SVM_test.py --data_path ./data/dev.pkl --feature_type mfcc

python3 SVM_test.py --data_path ./data/eval.pkl --feature_type mfcc

Referecne

[1] Ensemble Models for Spoofing Detection in Automatic Speaker Verification, 2019 Interspeech.

[2] IIIT-H Spoofing Countermeasures for Automatic Speaker Verification, 2019 Interspeech.

stoneMo/ASVspoof