wav2vec2-base-STTTest

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: 0.5198
• Wer: 0.3393

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
0.231	1.0	500	0.4337	0.4100
0.1845	2.01	1000	0.4296	0.3931
0.1551	3.01	1500	0.4397	0.3770
0.1479	4.02	2000	0.4524	0.3827
0.1186	5.02	2500	0.5182	0.3795
0.1079	6.02	3000	0.4799	0.3737
0.0974	7.03	3500	0.4966	0.3860
0.0878	8.03	4000	0.4993	0.3699
0.0788	9.04	4500	0.5183	0.3678
0.0732	10.04	5000	0.5064	0.3635
0.0664	11.04	5500	0.5330	0.3663
0.0596	12.05	6000	0.5147	0.3516
0.0538	13.05	6500	0.5254	0.3581
0.0535	14.06	7000	0.4902	0.3534
0.0492	15.06	7500	0.5115	0.3488
0.0455	16.06	8000	0.5250	0.3472
0.0434	17.07	8500	0.5338	0.3515
0.0351	18.07	9000	0.5365	0.3444
0.0341	19.08	9500	0.4886	0.3439
0.0332	20.08	10000	0.5234	0.3475
0.0289	21.08	10500	0.5375	0.3464
0.028	22.09	11000	0.5395	0.3478
0.0225	23.09	11500	0.5236	0.3428
0.0244	24.1	12000	0.5122	0.3402
0.0246	25.1	12500	0.5212	0.3390
0.0214	26.1	13000	0.5198	0.3393
0.0179	27.11	13500	0.5198	0.3393
0.0194	28.11	14000	0.5198	0.3393
0.0193	29.12	14500	0.5198	0.3393

Framework versions

• Transformers 4.17.0
• Pytorch 1.10.1+cu111
• Datasets 1.18.3
• Tokenizers 0.12.1