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: 0.5480
• Wer: 0.3437

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
3.5237	1.0	500	1.7277	0.9752
0.8339	2.01	1000	0.5413	0.5316
0.4277	3.01	1500	0.4732	0.4754
0.2907	4.02	2000	0.4571	0.4476
0.2254	5.02	2500	0.4611	0.4105
0.1911	6.02	3000	0.4448	0.4072
0.1595	7.03	3500	0.4517	0.3843
0.1377	8.03	4000	0.4551	0.3881
0.1197	9.04	4500	0.4853	0.3772
0.1049	10.04	5000	0.4617	0.3707
0.097	11.04	5500	0.4633	0.3622
0.0872	12.05	6000	0.4635	0.3690
0.0797	13.05	6500	0.5196	0.3749
0.0731	14.06	7000	0.5029	0.3639
0.0667	15.06	7500	0.5053	0.3614
0.0618	16.06	8000	0.5627	0.3638
0.0562	17.07	8500	0.5484	0.3577
0.0567	18.07	9000	0.5163	0.3560
0.0452	19.08	9500	0.5012	0.3538
0.044	20.08	10000	0.4931	0.3534
0.0424	21.08	10500	0.5147	0.3519
0.0356	22.09	11000	0.5540	0.3521
0.0322	23.09	11500	0.5565	0.3509
0.0333	24.1	12000	0.5315	0.3428
0.0281	25.1	12500	0.5284	0.3425
0.0261	26.1	13000	0.5101	0.3446
0.0256	27.11	13500	0.5432	0.3415
0.0229	28.11	14000	0.5484	0.3446
0.0212	29.12	14500	0.5480	0.3437

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1