whisper-small-ta-emotion
This model is a fine-tuned version of openai/whisper-small on the EmoTa (TamilSER-DB) dataset. It achieves the following results on the evaluation set:
- Loss: 3.0859
- Accuracy: 0.2556
- F1 Macro: 0.2109
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: 0.0005
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 0.1
- num_epochs: 12
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 Macro |
|---|---|---|---|---|---|
| No log | 1.0 | 24 | 3.3423 | 0.1944 | 0.0664 |
| 3.2189 | 2.0 | 48 | 3.2316 | 0.2444 | 0.1409 |
| 3.1391 | 3.0 | 72 | 3.1735 | 0.2333 | 0.1118 |
| 3.0224 | 4.0 | 96 | 3.1014 | 0.2556 | 0.1987 |
| 2.9275 | 5.0 | 120 | 3.2727 | 0.1889 | 0.1405 |
| 2.8341 | 6.0 | 144 | 3.0972 | 0.2611 | 0.1392 |
| 2.7619 | 7.0 | 168 | 3.0267 | 0.2 | 0.1597 |
| 2.7071 | 8.0 | 192 | 3.0764 | 0.2444 | 0.2072 |
| 2.6381 | 9.0 | 216 | 3.0706 | 0.2278 | 0.1613 |
| 2.5599 | 10.0 | 240 | 3.1087 | 0.25 | 0.1716 |
| 2.5374 | 11.0 | 264 | 3.0905 | 0.2611 | 0.2099 |
| 2.4927 | 12.0 | 288 | 3.0859 | 0.2556 | 0.2109 |
Framework versions
- Transformers 5.12.1
- Pytorch 2.10.0+cu128
- Datasets 5.0.0
- Tokenizers 0.22.2