wav2vec2-E50_freq_pause

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: 1.8196
• Cer: 36.9713

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
31.297	0.1289	200	5.4260	100.0
4.888	0.2579	400	4.7009	100.0
4.776	0.3868	600	5.0539	100.0
4.7621	0.5158	800	4.6373	100.0
4.6795	0.6447	1000	4.6474	100.0
4.6648	0.7737	1200	4.8580	100.0
4.6503	0.9026	1400	4.6060	100.0
4.5806	1.0316	1600	4.6559	98.0204
4.5544	1.1605	1800	4.4803	98.0909
4.4936	1.2895	2000	4.6766	98.0028
4.3834	1.4184	2200	4.3482	97.9088
4.2361	1.5474	2400	4.1447	80.8036
3.6904	1.6763	2600	3.3525	59.5277
3.1158	1.8053	2800	2.9366	53.3541
2.7611	1.9342	3000	2.6343	48.2202
2.527	2.0632	3200	2.3777	44.1788
2.323	2.1921	3400	2.1629	41.8351
2.18	2.3211	3600	2.1591	42.4695
2.0783	2.4500	3800	2.0168	39.9847
1.996	2.5790	4000	1.9937	40.1492
1.9336	2.7079	4200	1.8950	38.0815
1.8621	2.8369	4400	1.8627	37.7702
1.8707	2.9658	4600	1.8196	36.9713

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1