wav2vec2-base960-english-phoneme_v2
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: 0.4069
- Cer: 0.0900
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 |
|---|---|---|---|---|
| 2.18 | 6.94 | 500 | 0.3118 | 0.0923 |
| 0.2622 | 13.88 | 1000 | 0.4387 | 0.1218 |
| 0.2145 | 20.83 | 1500 | 0.4441 | 0.1121 |
| 0.1429 | 27.77 | 2000 | 0.4001 | 0.1045 |
| 0.0927 | 34.72 | 2500 | 0.4692 | 0.1062 |
| 0.0598 | 41.66 | 3000 | 0.3960 | 0.0971 |
| 0.0356 | 48.61 | 3500 | 0.4069 | 0.0900 |
Framework versions
- Transformers 4.23.0.dev0
- Pytorch 1.12.1.post201
- Datasets 2.5.2.dev0
- Tokenizers 0.12.1