wav2vec-arabic-V2-10

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: 0.3686
• Wer: 0.0763
• Per: 0.0613

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
22.0067	1.0	164	5.5136	1.0	1.0
3.7377	2.0	328	3.2749	1.0	1.0
3.2647	3.0	492	3.3163	1.0	1.0
3.2513	4.0	656	3.2319	1.0	1.0
3.2314	5.0	820	3.2902	1.0	1.0
3.1754	6.0	984	3.0334	1.0	1.0
2.8137	7.0	1148	2.5538	0.9930	0.9959
2.1146	8.0	1312	1.6577	0.6542	0.6458
1.2679	9.0	1476	0.8287	0.2089	0.1803
0.6981	10.0	1640	0.5801	0.1423	0.1229
0.4737	11.0	1804	0.4867	0.1219	0.1017
0.3733	12.0	1968	0.4482	0.1098	0.0929
0.2913	13.0	2132	0.4176	0.1092	0.0933
0.2358	14.0	2296	0.4200	0.1084	0.0915
0.2196	15.0	2460	0.3999	0.0987	0.0824
0.1845	16.0	2624	0.3993	0.1006	0.0826
0.1831	17.0	2788	0.3812	0.0810	0.0653
0.1629	18.0	2952	0.3676	0.0785	0.0636
0.1578	19.0	3116	0.3679	0.0806	0.0658
0.1317	20.0	3280	0.3686	0.0763	0.0613

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0