wav2vec_arabic_mdd

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3501
• Wer: 0.0432
• Per: 0.0327

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: 0.0001
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Per
8.1087	1.0	1637	3.1701	1.0	1.0
1.1869	2.0	3274	0.4045	0.0800	0.0646
0.2	3.0	4911	0.3260	0.0591	0.0464
0.1283	4.0	6548	0.3042	0.0618	0.0475
0.1037	5.0	8185	0.2727	0.0531	0.0410
0.0844	6.0	9822	0.3184	0.0543	0.0409
0.0738	7.0	11459	0.2886	0.0485	0.0366
0.0613	8.0	13096	0.3345	0.0488	0.0374
0.0573	9.0	14733	0.3469	0.0505	0.0394
0.0514	10.0	16370	0.3245	0.0510	0.0386
0.0469	11.0	18007	0.3094	0.0492	0.0374
0.0375	12.0	19644	0.3656	0.0521	0.0392
0.0356	13.0	21281	0.3296	0.0472	0.0356
0.0291	14.0	22918	0.3301	0.0448	0.0336
0.0242	15.0	24555	0.3575	0.0460	0.0357
0.0216	16.0	26192	0.3376	0.0443	0.0335
0.0208	17.0	27829	0.3688	0.0436	0.0332
0.018	18.0	29466	0.3673	0.0445	0.0340
0.0164	19.0	31103	0.3576	0.0432	0.0327
0.0128	20.0	32740	0.3501	0.0432	0.0327

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.0+cu118
• Datasets 2.14.6
• Tokenizers 0.14.1