wav2vec2-xlsr-persian-50p

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6846
• Wer: 0.4339

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.05	250	3.2104	1.0
3.2437	2.11	500	2.9131	1.0
3.2437	3.16	750	1.0335	0.7303
1.4382	4.22	1000	0.8335	0.6155
1.4382	5.27	1250	0.7640	0.5904
0.6923	6.33	1500	0.6923	0.5468
0.6923	7.38	1750	0.6627	0.5238
0.5137	8.44	2000	0.6606	0.5112
0.5137	9.49	2250	0.6600	0.5125
0.4258	10.55	2500	0.6337	0.4939
0.4258	11.6	2750	0.6454	0.4851
0.362	12.66	3000	0.6481	0.4793
0.362	13.71	3250	0.6487	0.4801
0.3179	14.77	3500	0.6602	0.4668
0.3179	15.82	3750	0.6757	0.4683
0.2861	16.88	4000	0.6544	0.4591
0.2861	17.93	4250	0.6659	0.4634
0.2529	18.99	4500	0.6311	0.4556
0.2529	20.04	4750	0.6574	0.4525
0.235	21.1	5000	0.7019	0.4462
0.235	22.15	5250	0.6783	0.4426
0.2203	23.21	5500	0.6789	0.4361
0.2203	24.26	5750	0.6779	0.4336
0.2014	25.32	6000	0.6805	0.4406
0.2014	26.37	6250	0.6918	0.4407
0.1957	27.43	6500	0.6919	0.4360
0.1957	28.48	6750	0.6795	0.4332
0.1837	29.53	7000	0.6846	0.4339

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu113
• Datasets 1.18.3
• Tokenizers 0.10.3