base Turkish Whisper (bTW)
This model is a fine-tuned version of openai/whisper-base on the Ermetal Meetings dataset. It achieves the following results on the evaluation set:
- Loss: 1.9500
- Wer: 2.1895
- Cer: 1.3548
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: 1e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 4
- 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: 500
- training_steps: 1000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 1.7116 | 5.53 | 100 | 1.9115 | 1.1785 | 0.6901 |
| 0.6101 | 11.11 | 200 | 1.5123 | 1.1039 | 0.6221 |
| 0.2376 | 16.64 | 300 | 1.5636 | 0.9817 | 0.6448 |
| 0.0591 | 22.21 | 400 | 1.7179 | 2.2005 | 1.3384 |
| 0.0177 | 27.75 | 500 | 1.8454 | 1.9205 | 1.2140 |
| 0.0096 | 33.32 | 600 | 1.8529 | 1.2983 | 0.7777 |
| 0.0048 | 38.85 | 700 | 1.9306 | 2.3411 | 1.4385 |
| 0.0032 | 44.43 | 800 | 1.9388 | 1.9523 | 1.2705 |
| 0.0028 | 49.96 | 900 | 1.9472 | 1.8655 | 1.2023 |
| 0.0026 | 55.53 | 1000 | 1.9500 | 2.1895 | 1.3548 |
Framework versions
- Transformers 4.26.0
- Pytorch 1.12.0+cu102
- Datasets 2.9.0
- Tokenizers 0.13.2