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: 0.5195
• Wer: 0.3418

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
3.5159	1.0	500	1.7153	1.0291
0.8502	2.01	1000	0.5204	0.5146
0.431	3.01	1500	0.4491	0.4537
0.3073	4.02	2000	0.3883	0.4190
0.2338	5.02	2500	0.4453	0.4230
0.1956	6.02	3000	0.4599	0.3981
0.1594	7.03	3500	0.4240	0.3916
0.1423	8.03	4000	0.4756	0.3975
0.1252	9.04	4500	0.4427	0.3827
0.1064	10.04	5000	0.4489	0.3809
0.101	11.04	5500	0.4531	0.3961
0.0877	12.05	6000	0.4881	0.3883
0.0817	13.05	6500	0.5023	0.3774
0.0703	14.06	7000	0.5078	0.3679
0.0663	15.06	7500	0.5279	0.3620
0.0584	16.06	8000	0.5112	0.3653
0.0579	17.07	8500	0.4959	0.3633
0.0572	18.07	9000	0.4676	0.3626
0.0502	19.08	9500	0.5216	0.3503
0.0432	20.08	10000	0.4946	0.3480
0.0417	21.08	10500	0.4949	0.3532
0.0335	22.09	11000	0.5485	0.3557
0.032	23.09	11500	0.5087	0.3464
0.0334	24.1	12000	0.5313	0.3498
0.0263	25.1	12500	0.5148	0.3457
0.0242	26.1	13000	0.5232	0.3442
0.0235	27.11	13500	0.5122	0.3418
0.0221	28.11	14000	0.5074	0.3407
0.0215	29.12	14500	0.5195	0.3418

Framework versions

• Transformers 4.24.0
• Pytorch 1.13.0+cu116
• Datasets 1.18.3
• Tokenizers 0.13.2