wav2vec2-E30_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: 1.0467
• Cer: 28.3130

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
28.57	0.1289	200	4.9399	100.0
4.9152	0.2579	400	4.7298	100.0
4.7776	0.3868	600	4.6311	98.1732
4.7311	0.5158	800	4.5605	97.6739
4.6426	0.6447	1000	4.5556	97.7032
4.5691	0.7737	1200	4.5028	97.4330
4.1847	0.9026	1400	3.9048	81.7375
3.1837	1.0316	1600	2.8792	57.0724
2.6116	1.1605	1800	2.4695	49.7827
2.2803	1.2895	2000	2.2168	43.7559
2.0438	1.4184	2200	1.9216	40.6074
1.8919	1.5474	2400	1.7582	39.0273
1.7295	1.6763	2600	1.6734	38.5103
1.5832	1.8053	2800	1.5192	34.3221
1.4426	1.9342	3000	1.4440	33.6642
1.3355	2.0632	3200	1.3543	33.4821
1.2131	2.1921	3400	1.2427	31.7669
1.1532	2.3211	3600	1.2136	31.8785
1.0948	2.4500	3800	1.1645	30.5804
1.0283	2.5790	4000	1.1471	29.8931
1.0085	2.7079	4200	1.0822	28.8181
0.9753	2.8369	4400	1.0493	28.3306
0.976	2.9658	4600	1.0467	28.3130

Framework versions

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