wav2vec2-base-cynthia-tedlium-2500
This model is a fine-tuned version of facebook/wav2vec2-base-960h on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4356
- Wer: 0.1796
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
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- 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 |
|---|---|---|---|---|
| 6.2811 | 6.58 | 500 | 3.1455 | 0.9934 |
| 3.0533 | 13.16 | 1000 | 2.9568 | 0.9934 |
| 1.8269 | 19.73 | 1500 | 0.7595 | 0.2484 |
| 0.5103 | 26.31 | 2000 | 0.4356 | 0.1796 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu113
- Datasets 1.13.3
- Tokenizers 0.10.3