wav2vec2_ASV_deepfake_audio_detection_DF_finetune_frozen_further

This model is a fine-tuned version of Bisher/wav2vec2_ASV_deepfake_audio_detection on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3702
• Accuracy: 0.9173
• Precision: 0.9193
• Recall: 0.9173
• F1: 0.8917
• Tp: 384
• Tn: 17889
• Fn: 1623
• Fp: 24
• Eer: 0.0762
• Min Tdcf: 0.0323
• Auc Roc: 0.9561

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 152
• eval_batch_size: 152
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 608
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1	Tp	Tn	Fn	Fp	Eer	Min Tdcf	Auc Roc
0.0237	0.0490	10	0.3313	0.9162	0.9174	0.9162	0.8897	364	17886	1643	27	0.0829	0.0335	0.9725
0.0219	0.0979	20	0.3291	0.9147	0.9168	0.9147	0.8865	328	17892	1679	21	0.0809	0.0328	0.9736
0.0202	0.1469	30	0.3233	0.9173	0.9189	0.9173	0.8918	386	17887	1621	26	0.0765	0.0323	0.9738
0.0205	0.1958	40	0.3702	0.9173	0.9193	0.9173	0.8917	384	17889	1623	24	0.0762	0.0323	0.9561

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0
• Datasets 2.21.0
• Tokenizers 0.19.1