wav2vec2_ASV_deepfake_audio_detection_DF_finetune_frozen_chngd_classifier

This model is a fine-tuned version of Bisher/wav2vec2_ASV_deepfake_audio_detection_DF_finetune_frozen on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.6355
Accuracy: 0.9146
Precision: 0.9161
Recall: 0.9146
F1: 0.8866
Tp: 330
Tn: 17889
Fn: 1677
Fp: 24
Eer: 0.1639
Min Tdcf: 0.0357
Auc Roc: 0.9189

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: 0.0005
train_batch_size: 128
eval_batch_size: 128
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 512
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
0.6678	0.0206	5	0.6581	0.9410	0.9381	0.9410	0.9331	963	17782	1044	131	0.0717	0.0322	0.9729
0.6124	0.0412	10	0.5702	0.9060	0.9082	0.9060	0.8681	145	17902	1862	11	0.1978	0.0322	0.8640
0.5335	0.0619	15	0.5016	0.9016	0.9093	0.9016	0.8573	48	17912	1959	1	0.3503	0.0363	0.6924
0.4592	0.0825	20	0.4335	0.9094	0.9108	0.9094	0.8759	221	17895	1786	18	0.1774	0.0323	0.8683
0.3927	0.1031	25	0.3781	0.9104	0.9110	0.9104	0.8782	244	17891	1763	22	0.2297	0.0315	0.8110
0.3231	0.1237	30	0.3201	0.9138	0.9151	0.9138	0.8851	314	17889	1693	24	0.2900	0.0309	0.7424
0.2519	0.1443	35	0.2804	0.9196	0.9215	0.9196	0.8960	431	17887	1576	26	0.1141	0.0295	0.9340
0.1963	0.1649	40	0.2395	0.9346	0.9350	0.9346	0.9216	751	17866	1256	47	0.0898	0.0286	0.9623
0.1423	0.1856	45	0.3794	0.9048	0.9032	0.9048	0.8659	127	17897	1880	16	0.0901	0.0298	0.9659
0.1046	0.2062	50	0.3194	0.9287	0.9286	0.9287	0.9124	636	17863	1371	50	0.0751	0.0318	0.9767
0.0681	0.2268	55	0.4859	0.9021	0.9055	0.9021	0.8586	60	17909	1947	4	0.1709	0.0378	0.9015
0.0473	0.2474	60	0.5605	0.9100	0.9101	0.9100	0.8774	237	17890	1770	23	0.7055	0.0382	0.3149
0.0323	0.2680	65	0.5107	0.9164	0.9178	0.9164	0.8900	367	17887	1640	26	0.0703	0.0337	0.9791
0.0339	0.2887	70	0.8921	0.9026	0.9009	0.9026	0.8604	77	17903	1930	10	0.8316	0.0435	0.1773
0.0423	0.3093	75	0.8964	0.9030	0.8998	0.9030	0.8615	87	17900	1920	13	0.0732	0.0327	0.9753
0.0456	0.3299	80	1.0843	0.9013	0.8935	0.9013	0.8574	51	17902	1956	11	0.8520	0.0478	0.1126
0.0712	0.3505	85	0.8587	0.9023	0.8998	0.9023	0.8597	71	17903	1936	10	0.8665	0.0480	0.0990
0.0629	0.3711	90	0.4810	0.9267	0.9278	0.9267	0.9087	583	17877	1424	36	0.0848	0.0328	0.9683
0.0477	0.3918	95	0.9415	0.9094	0.9114	0.9094	0.8757	218	17897	1789	16	0.1219	0.0408	0.8890
0.0484	0.4124	100	0.7774	0.9150	0.9170	0.9150	0.8873	336	17891	1671	22	0.6906	0.0383	0.3129
0.0449	0.4330	105	0.3949	0.9197	0.9199	0.9197	0.8967	444	17876	1563	37	0.6527	0.0363	0.3629
0.0567	0.4536	110	0.5853	0.9232	0.9212	0.9232	0.9040	540	17850	1467	63	0.2192	0.0355	0.8158
0.0416	0.4742	115	0.7031	0.9036	0.9054	0.9036	0.8626	95	17905	1912	8	0.7633	0.0408	0.2549
0.1778	0.4948	120	0.5440	0.9033	0.9093	0.9033	0.8613	83	17910	1924	3	0.7389	0.0398	0.2838
0.036	0.5155	125	0.5825	0.9161	0.9187	0.9161	0.8892	356	17893	1651	20	0.1538	0.0366	0.9078
0.0797	0.5361	130	0.5864	0.9027	0.9059	0.9027	0.8601	73	17908	1934	5	0.2236	0.0407	0.7721
0.0669	0.5567	135	0.4264	0.9036	0.9079	0.9036	0.8623	92	17908	1915	5	0.1597	0.0370	0.8791
0.0353	0.5773	140	0.6355	0.9146	0.9161	0.9146	0.8866	330	17889	1677	24	0.1639	0.0357	0.9189

Framework versions

Transformers 4.44.0
Pytorch 2.4.0
Datasets 2.21.0
Tokenizers 0.19.1

Bisher/wav2vec2_ASV_deepfake_audio_detection_DF_finetune_frozen_chngd_classifier

wav2vec2_ASV_deepfake_audio_detection_DF_finetune_frozen_chngd_classifier

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

为此模型做种