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: 1.1836
- Wer: 1.7109
- Cer: 1.2860
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 |
|---|---|---|---|---|---|
| 1.7878 | 4.74 | 100 | 1.4516 | 0.8560 | 0.5525 |
| 0.6701 | 9.51 | 200 | 0.9194 | 0.8543 | 0.6112 |
| 0.3364 | 14.28 | 300 | 0.8871 | 0.7415 | 0.4992 |
| 0.1228 | 19.05 | 400 | 0.9671 | 0.9052 | 0.6678 |
| 0.0355 | 23.78 | 500 | 1.0515 | 0.8961 | 0.6208 |
| 0.0148 | 28.55 | 600 | 1.0684 | 0.6644 | 0.3694 |
| 0.0056 | 33.32 | 700 | 1.1488 | 1.3315 | 0.8732 |
| 0.0041 | 38.09 | 800 | 1.1700 | 1.7415 | 1.1934 |
| 0.0034 | 42.83 | 900 | 1.1801 | 1.7745 | 1.2643 |
| 0.0032 | 47.6 | 1000 | 1.1836 | 1.7109 | 1.2860 |
Framework versions
- Transformers 4.26.0
- Pytorch 1.12.0+cu102
- Datasets 2.9.0
- Tokenizers 0.13.2