wav2vec2-base-timit-eng

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

• Loss: 0.5047
• Wer: 0.2233

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
3.5485	1.0	500	1.9954	1.0042
0.9068	2.01	1000	0.6418	0.4572
0.4398	3.01	1500	0.4586	0.3629
0.3023	4.02	2000	0.4464	0.3248
0.2328	5.02	2500	0.4019	0.2969
0.1899	6.02	3000	0.4363	0.2961
0.163	7.03	3500	0.4832	0.2872
0.1442	8.03	4000	0.4421	0.2801
0.1246	9.04	4500	0.4757	0.2659
0.1122	10.04	5000	0.4693	0.2648
0.102	11.04	5500	0.4834	0.2549
0.0919	12.05	6000	0.4558	0.2633
0.0866	13.05	6500	0.4527	0.2641
0.0762	14.06	7000	0.4394	0.2565
0.0705	15.06	7500	0.5240	0.2609
0.0647	16.06	8000	0.4980	0.2522
0.0608	17.07	8500	0.5163	0.2589
0.0576	18.07	9000	0.4991	0.2565
0.0499	19.08	9500	0.4750	0.2457
0.047	20.08	10000	0.5162	0.2447
0.0418	21.08	10500	0.4801	0.2413
0.0383	22.09	11000	0.4961	0.2394
0.0342	23.09	11500	0.5209	0.2386
0.032	24.1	12000	0.4970	0.2364
0.0293	25.1	12500	0.4789	0.2309
0.0265	26.1	13000	0.4948	0.2302
0.0269	27.11	13500	0.4917	0.2249
0.0237	28.11	14000	0.4991	0.2238
0.022	29.12	14500	0.5047	0.2233

Framework versions

• Transformers 4.17.0
• Pytorch 1.13.0+cu116
• Datasets 1.18.3
• Tokenizers 0.13.2