wav2vec2-large-english-phoneme
This model is a fine-tuned version of facebook/wav2vec2-large on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3463
- Cer: 0.1086
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: 0.0003
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 50
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 1.6406 | 6.94 | 500 | 0.3054 | 0.1201 |
| 0.3534 | 13.88 | 1000 | 0.3461 | 0.1294 |
| 0.3142 | 20.83 | 1500 | 0.3163 | 0.1211 |
| 0.2501 | 27.77 | 2000 | 0.3119 | 0.1137 |
| 0.1993 | 34.72 | 2500 | 0.3157 | 0.1133 |
| 0.1557 | 41.66 | 3000 | 0.3402 | 0.1113 |
| 0.1191 | 48.61 | 3500 | 0.3463 | 0.1086 |
Framework versions
- Transformers 4.21.3
- Pytorch 1.12.1.post200
- Datasets 2.4.0
- Tokenizers 0.12.1