wav2vec2-base-timit-demo-colab12
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: 0.4831
- Wer: 0.3546
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: 420
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 4.1683 | 3.52 | 500 | 1.3684 | 0.7364 |
| 0.7614 | 7.04 | 1000 | 0.6008 | 0.5218 |
| 0.4721 | 10.56 | 1500 | 0.5319 | 0.4614 |
| 0.3376 | 14.08 | 2000 | 0.5234 | 0.4308 |
| 0.2508 | 17.61 | 2500 | 0.5109 | 0.3998 |
| 0.1978 | 21.13 | 3000 | 0.5037 | 0.3721 |
| 0.1645 | 24.65 | 3500 | 0.4918 | 0.3622 |
| 0.1449 | 28.17 | 4000 | 0.4831 | 0.3546 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.11.0+cu113
- Datasets 1.18.3
- Tokenizers 0.10.3