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: 0.5320
• Wer: 0.3362

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
3.5752	1.0	500	2.2162	1.0392
0.9682	2.01	1000	0.5208	0.5191
0.4414	3.01	1500	0.4623	0.4528
0.2971	4.02	2000	0.4553	0.4421
0.2328	5.02	2500	0.4960	0.4208
0.1891	6.02	3000	0.4213	0.3812
0.1571	7.03	3500	0.4863	0.3865
0.1544	8.03	4000	0.4677	0.3910
0.1302	9.04	4500	0.6305	0.4042
0.1111	10.04	5000	0.5104	0.3830
0.0986	11.04	5500	0.5332	0.3808
0.088	12.05	6000	0.4494	0.3674
0.0827	13.05	6500	0.4779	0.3748
0.074	14.06	7000	0.5315	0.3738
0.0664	15.06	7500	0.5367	0.3661
0.0568	16.06	8000	0.5707	0.3817
0.057	17.07	8500	0.5381	0.3719
0.0561	18.07	9000	0.5353	0.3705
0.0487	19.08	9500	0.5087	0.3579
0.0444	20.08	10000	0.4910	0.3596
0.0433	21.08	10500	0.4931	0.3497
0.0363	22.09	11000	0.5414	0.3488
0.0318	23.09	11500	0.5405	0.3472
0.033	24.1	12000	0.5476	0.3449
0.0262	25.1	12500	0.5529	0.3443
0.0255	26.1	13000	0.5299	0.3417
0.0252	27.11	13500	0.5092	0.3363
0.0221	28.11	14000	0.5309	0.3357
0.0223	29.12	14500	0.5320	0.3362

Framework versions

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