wav2vec2-xls-r-300m-ar

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

• Loss: 0.3212
• Wer: 0.0636

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
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0793	8.85	1000	0.1626	0.0786
0.0396	17.7	2000	0.2199	0.0807
0.0285	26.55	3000	0.2289	0.0694
0.021	35.4	4000	0.2662	0.0722
0.0177	44.25	5000	0.2459	0.0744
0.0155	53.1	6000	0.2689	0.0679
0.0149	61.95	7000	0.2760	0.0717
0.0074	70.8	8000	0.3004	0.0680
0.0058	79.65	9000	0.3113	0.0650
0.0033	88.5	10000	0.3212	0.0636

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.11.1.dev0
• Tokenizers 0.13.2