whisper-small-fa
This model is a fine-tuned version of openai/whisper-small on an unknown dataset. This was an experiment; better results are likely with more data and longer training. It achieves the following results on the evaluation set:
- Loss: 0.1537
- Wer: 19.2460
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 8
- 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: 500
- training_steps: 4000
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.2216 | 0.1935 | 1000 | 0.2209 | 28.1653 |
| 0.1947 | 0.3871 | 2000 | 0.1808 | 24.9731 |
| 0.1465 | 0.5806 | 3000 | 0.1621 | 20.7613 |
| 0.129 | 0.7741 | 4000 | 0.1537 | 19.2460 |
Note: Early stopping at 4k steps due to rising gap (train vs val) indicating overfitting.
How to use
from transformers import pipeline
asr = pipeline(
task="automatic-speech-recognition",
model="kiarashQ/fa-ir-stt-whisper-small-v1",
chunk_length_s=30,
stride_length_s=(5, 5),
return_timestamps=False
)
out = asr("example.wav")
print(out["text"])
Framework versions
- Transformers 4.56.2
- Pytorch 2.8.0+cu128
- Datasets 4.1.1
- Tokenizers 0.22.1