metadata
language:
- tr
license: apache-2.0
tags:
- hf-asr-leaderboard
- generated_from_trainer
metrics:
- wer
model-index:
- name: base Turkish Whisper (bTW)
results: []
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: 1.8564
- Wer: 1.2482
- Cer: 0.7381
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 |
|---|---|---|---|---|---|
| 1.6604 | 2.86 | 100 | 1.9378 | 1.1296 | 0.6334 |
| 0.6453 | 5.71 | 200 | 1.4655 | 0.9878 | 0.5974 |
| 0.3912 | 8.57 | 300 | 1.4669 | 1.2543 | 0.7557 |
| 0.2081 | 11.43 | 400 | 1.4622 | 0.8203 | 0.5123 |
| 0.094 | 14.29 | 500 | 1.6592 | 0.9535 | 0.6367 |
| 0.039 | 17.14 | 600 | 1.6946 | 0.9658 | 0.5706 |
| 0.0172 | 20.0 | 700 | 1.8271 | 1.4046 | 1.0027 |
| 0.0086 | 22.86 | 800 | 1.8149 | 1.2567 | 0.7530 |
| 0.0064 | 25.71 | 900 | 1.8478 | 1.2311 | 0.7279 |
| 0.0061 | 28.57 | 1000 | 1.8564 | 1.2482 | 0.7381 |
Framework versions
- Transformers 4.26.0
- Pytorch 1.12.0+cu102
- Datasets 2.9.0
- Tokenizers 0.13.2