CustomWav2Vec2ForCTC

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0944
• Wer: 0.1712

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: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1786	1.05	500	0.1349	0.2173
0.3532	2.11	1000	0.2127	0.2367
0.3307	3.16	1500	0.4717	0.2201
0.2932	4.21	2000	0.1808	0.2388
0.2591	5.26	2500	0.1754	0.2277
0.2512	6.32	3000	0.5620	0.2551
0.2205	7.37	3500	0.1447	0.2137
0.2135	8.42	4000	0.1434	0.2010
0.1815	9.47	4500	0.1127	0.1912
0.1794	10.53	5000	0.2051	0.2229
0.1587	11.58	5500	0.1050	0.1934
0.1587	12.63	6000	0.1322	0.2036
0.1325	13.68	6500	0.1309	0.1996
0.1337	14.74	7000	0.1110	0.1872
0.1258	15.79	7500	0.1009	0.1811
0.1132	16.84	8000	0.1083	0.1882
0.1077	17.89	8500	0.1043	0.1813
0.1055	18.95	9000	0.1011	0.1810
0.0923	20.0	9500	0.1140	0.1809
0.0899	21.05	10000	0.1298	0.1823
0.0874	22.11	10500	0.1064	0.1735
0.0745	23.16	11000	0.1046	0.1752
0.0726	24.21	11500	0.0977	0.1760
0.066	25.26	12000	0.1018	0.1745
0.0646	26.32	12500	0.0990	0.1713
0.0564	27.37	13000	0.0961	0.1709
0.0622	28.42	13500	0.0935	0.1717
0.0526	29.47	14000	0.0944	0.1712

Framework versions

• Transformers 4.17.0
• Pytorch 2.5.1+cu121
• Datasets 1.18.3
• Tokenizers 0.20.3