wav2vec2-vivos-asr
This model is a fine-tuned version of facebook/wav2vec2-base on the vivos dataset. It achieves the following results on the evaluation set:
- Loss: 0.6926
- Wer: 0.4232
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: 0.0002
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 300
- num_epochs: 20
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 6.3715 | 2.0 | 146 | 3.6727 | 1.0 |
| 3.4482 | 4.0 | 292 | 3.5947 | 1.0 |
| 3.4187 | 6.0 | 438 | 3.5349 | 1.0 |
| 3.3922 | 8.0 | 584 | 3.4713 | 1.0 |
| 3.349 | 10.0 | 730 | 3.3434 | 1.0 |
| 2.1445 | 12.0 | 876 | 1.3684 | 0.7849 |
| 1.0296 | 14.0 | 1022 | 0.9135 | 0.5588 |
| 0.7796 | 16.0 | 1168 | 0.7838 | 0.4871 |
| 0.609 | 18.0 | 1314 | 0.7060 | 0.4372 |
| 0.5388 | 20.0 | 1460 | 0.6926 | 0.4232 |
Framework versions
- Transformers 4.44.0
- Pytorch 2.4.0
- Datasets 2.21.0
- Tokenizers 0.19.1