wav2vec2-large-xlsr-coraa-aug-texts-exp-1

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2603
• Wer: 0.1877
• Cer: 0.1189
• Per: 0.1368

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: 3e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Per
0.0959	1.0	64	0.2713	0.1881	0.1202	0.1385
0.1109	2.0	128	0.2630	0.1913	0.1188	0.1358
0.1109	3.0	192	0.2906	0.1917	0.1208	0.1383
0.1099	4.0	256	0.2603	0.1877	0.1189	0.1368
0.0986	5.0	320	0.2733	0.1889	0.1192	0.1364
0.0986	6.0	384	0.2649	0.1868	0.1191	0.1360
0.0949	7.0	448	0.2987	0.1935	0.1223	0.1387
0.1037	8.0	512	0.2819	0.1868	0.1192	0.1361
0.1037	9.0	576	0.2867	0.1870	0.1183	0.1352
0.0857	10.0	640	0.2928	0.1836	0.1181	0.1352
0.0803	11.0	704	0.2896	0.1864	0.1191	0.1358
0.0803	12.0	768	0.2733	0.1799	0.1164	0.1333
0.0869	13.0	832	0.2648	0.1819	0.1172	0.1342
0.0869	14.0	896	0.2800	0.1860	0.1184	0.1353

Framework versions

• Transformers 4.28.0
• Pytorch 2.6.0+cu124
• Datasets 4.0.0
• Tokenizers 0.13.3