A modular, configuration-driven framework. Mix frontends, backends, and losses via YAML — no code changes needed.
Everything you need for deepfake audio detection research, in one framework.
Swap Wav2Vec2 for WavLM, or AASIST for MLP, with a single line of YAML. All components are decoupled and interchangeable.
Define model architecture, data pipeline, augmentations, loss functions, and training schedule in one YAML file.
Download any of 455+ pretrained models and 12 benchmark datasets with a single CLI command.
Automatic tracking of EER, minDCF, actDCF, F1, and Accuracy with WandB logging and checkpointing.
Built-in RawBoost, RIR, Codec, AdditiveNoise, SpeedPerturb, AddBabble, DropFreq, DropChunk, and more.
Configs are saved with every experiment. Same seed + same config = identical results.
All components are plug-and-play via the registry pattern. Use built-in ones or add your own.
Pretrained SSL feature extractors
Classifier heads
Training objectives
Data augmentation pipeline
Evaluation metrics
Optimization & scheduling
455+ pretrained models and 12 benchmark datasets, ready to download.
Checkpoints across 12 datasets, multiple frontends (WavLM, Wav2Vec2, EAT, MERT), backends (AASIST, Nes2Net, MLP, TCM), and seeds.
deepfense download model ASV19_WavLM_Nes2Net_NoAug_Seed42
ASVSpoof19, CompSpoof, DECRO, SONICS, SpeechFake, WaveFake, FakeMusicCaps, and more — as ready-to-use Parquet files.
deepfense download dataset CompSpoof
Search for models by frontend, backend, dataset, or seed from the CLI. Download the config and checkpoint in one command.
deepfense download list-models --filter WavLM
Install, download, and train in under 5 minutes.
pip install deepfensegit clone https://github.com/Yaselley/deepfense-framework
cd deepfense-framework
pip install -e .deepfense download dataset CompSpoof
deepfense download model ASV19_WavLM_Nes2Net_NoAug_Seed42deepfense train --config deepfense/config/train.yamldeepfense test \
--config models/ASV19_WavLM_Nes2Net_NoAug_Seed42/config.yaml \
--checkpoint models/ASV19_WavLM_Nes2Net_NoAug_Seed42/best_model.pthdeepfense listFull documentation with tutorials, config reference, and extension guides.
DeepFense is Apache 2.0 licensed. Contributions welcome — add new models, datasets, augmentations, or improvements.
Contribute on GitHub