Modelpublic

nrshoudi/wav2vec_arabic_mdd_v2

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes16downloads

Model Card

wav2vecarabicmdd_v2

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.2736
—Wer: 0.0492
—Cer: 0.0378

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: 0.0003
—trainbatchsize: 16
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 20
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
5.2969	0.9951	102	4.4152	1.0	1.0
3.2462	2.0	205	3.2917	1.0	1.0
3.1998	2.9951	307	3.2287	1.0	1.0
3.2577	4.0	410	3.1610	1.0	1.0
2.4548	4.9951	512	2.5563	0.9881	0.9914
0.678	6.0	615	0.7636	0.2986	0.2701
0.1777	6.9951	717	0.3790	0.0925	0.0781
0.1097	8.0	820	0.3732	0.0865	0.0694
0.0737	8.9951	922	0.3027	0.0641	0.0511
0.0526	10.0	1025	0.2834	0.0699	0.0578
0.0471	10.9951	1127	0.2601	0.0541	0.0435
0.0349	12.0	1230	0.2803	0.0518	0.0396
0.029	12.9951	1332	0.2710	0.0502	0.0378
0.0225	14.0	1435	0.2835	0.0494	0.0378
0.023	14.9951	1537	0.2909	0.0483	0.0368
0.0247	16.0	1640	0.2725	0.0480	0.0361
0.035	16.9951	1742	0.2696	0.0489	0.0372
0.0156	18.0	1845	0.2742	0.0482	0.0364
0.0183	18.9951	1947	0.2741	0.0492	0.0376
0.0179	19.9024	2040	0.2736	0.0492	0.0378

Framework versions

—Transformers 4.40.0
—Pytorch 2.2.1+cu121
—Datasets 2.19.0
—Tokenizers 0.19.1