<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
wav2vec2-base-timit-demo-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: 0.4234
- Wer: 0.2752
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: 16
- 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 |
|---|---|---|---|---|
| 3.5378 | 1.73 | 500 | 1.3767 | 0.8694 |
| 0.6656 | 3.46 | 1000 | 0.4546 | 0.3897 |
| 0.3113 | 5.19 | 1500 | 0.3904 | 0.3470 |
| 0.1922 | 6.92 | 2000 | 0.3746 | 0.3178 |
| 0.1433 | 8.65 | 2500 | 0.3837 | 0.3120 |
| 0.1146 | 10.38 | 3000 | 0.3520 | 0.3016 |
| 0.0946 | 12.11 | 3500 | 0.3713 | 0.3096 |
| 0.0793 | 13.84 | 4000 | 0.4005 | 0.3021 |
| 0.0667 | 15.57 | 4500 | 0.4141 | 0.2941 |
| 0.0565 | 17.3 | 5000 | 0.4006 | 0.2888 |
| 0.0514 | 19.03 | 5500 | 0.4050 | 0.2908 |
| 0.0466 | 20.76 | 6000 | 0.3882 | 0.2844 |
| 0.0387 | 22.49 | 6500 | 0.4054 | 0.2806 |
| 0.0337 | 24.22 | 7000 | 0.4094 | 0.2800 |
| 0.0312 | 25.95 | 7500 | 0.4291 | 0.2815 |
| 0.0278 | 27.68 | 8000 | 0.3999 | 0.2761 |
| 0.0251 | 29.41 | 8500 | 0.4234 | 0.2752 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0
- Datasets 1.13.3
- Tokenizers 0.10.3
