<!-- 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: 0.4545
- Wer: 0.3450
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: 32
- 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 |
|---|---|---|---|---|
| 3.3801 | 4.0 | 500 | 1.1501 | 0.8820 |
| 0.561 | 8.0 | 1000 | 0.4583 | 0.4211 |
| 0.2198 | 12.0 | 1500 | 0.4467 | 0.3997 |
| 0.1255 | 16.0 | 2000 | 0.4390 | 0.3677 |
| 0.0862 | 20.0 | 2500 | 0.4934 | 0.3603 |
| 0.0617 | 24.0 | 3000 | 0.4641 | 0.3549 |
| 0.0465 | 28.0 | 3500 | 0.4545 | 0.3450 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.10.3
