Whisper Small Indonesian
This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0, magic_data, titml, google/fleurs dataset. It achieves the following results on the evaluation set:
- Loss: 0.1022
- Wer: 6.0592
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: 32
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 10000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.173 | 0.66 | 1000 | 0.1654 | 9.8773 |
| 0.0771 | 1.32 | 2000 | 0.1290 | 7.7515 |
| 0.0569 | 1.99 | 3000 | 0.1056 | 7.1475 |
| 0.0274 | 2.65 | 4000 | 0.1044 | 6.6264 |
| 0.0072 | 3.31 | 5000 | 0.1023 | 6.3543 |
| 0.009 | 3.97 | 6000 | 0.1000 | 6.3359 |
| 0.0033 | 4.63 | 7000 | 0.1022 | 6.0592 |
| 0.002 | 5.29 | 8000 | 0.1051 | 6.1560 |
| 0.0028 | 5.96 | 9000 | 0.1052 | 6.1007 |
| 0.0013 | 6.62 | 10000 | 0.1063 | 6.1376 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.10.2+cu102
- Datasets 2.7.0
- Tokenizers 0.13.1