wav2vec2-base-timit-demo-google-colab

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: 0.5430
• Wer: 0.3434

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
3.5342	1.0	500	1.7432	1.0005
0.852	2.01	1000	0.6003	0.5692
0.4306	3.01	1500	0.4681	0.4750
0.2972	4.02	2000	0.4397	0.4192
0.2262	5.02	2500	0.4120	0.3985
0.1902	6.02	3000	0.4680	0.3988
0.1664	7.03	3500	0.4740	0.4022
0.1398	8.03	4000	0.4200	0.3788
0.1237	9.04	4500	0.4645	0.3827
0.1087	10.04	5000	0.4520	0.3805
0.0961	11.04	5500	0.4862	0.3712
0.0844	12.05	6000	0.4768	0.3695
0.0808	13.05	6500	0.5238	0.3720
0.0708	14.06	7000	0.5249	0.3736
0.0649	15.06	7500	0.5245	0.3745
0.0639	16.06	8000	0.5152	0.3648
0.0553	17.07	8500	0.5048	0.3682
0.0556	18.07	9000	0.5316	0.3634
0.0469	19.08	9500	0.5179	0.3668
0.0432	20.08	10000	0.5441	0.3679
0.0402	21.08	10500	0.5362	0.3498
0.0344	22.09	11000	0.5497	0.3537
0.0321	23.09	11500	0.5426	0.3499
0.0311	24.1	12000	0.5637	0.3508
0.0279	25.1	12500	0.5346	0.3491
0.025	26.1	13000	0.5310	0.3426
0.0237	27.11	13500	0.5419	0.3444
0.0221	28.11	14000	0.5457	0.3437
0.0214	29.12	14500	0.5430	0.3434

Framework versions

• Transformers 4.17.0
• Pytorch 1.12.1+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1