README.md

---
language:
- es
license: apache-2.0
base_model: openai/whisper-large-v2
tags:
- whisper-event
- generated_from_trainer
datasets:
- mozilla-foundation/common_voice_13_0
metrics:
- wer
model-index:
- name: Whisper Large-V2 Spanish
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: mozilla-foundation/common_voice_13_0 es
      type: mozilla-foundation/common_voice_13_0
      config: es
      split: test
      args: es
    metrics:
    - name: Wer
      type: wer
      value: 4.89488506963824
---

# Whisper Large-V2 Spanish

## Model summary

**Whisper Large-V2 Spanish** is a state-of-the-art automatic speech recognition (ASR) model for **Spanish (es)**, fine-tuned from [openai/whisper-large-v2] on the **Spanish subset of Mozilla Common Voice 13.0**. It achieves a **Word Error Rate (WER) of 4.8949%** on the evaluation set.

This model is optimized for applications that demand **very high transcription accuracy** while benefiting from improvements in Large-V2 architecture over the original Large model.

---

## Model description

* **Architecture:** Transformer-based encoder–decoder (Whisper Large-V2)  
* **Base model:** openai/whisper-large-v2  
* **Language:** Spanish (es)  
* **Task:** Automatic Speech Recognition (ASR)  
* **Output:** Text transcription in Spanish  
* **Decoding:** Autoregressive sequence-to-sequence decoding  

Large-V2 introduces enhanced pretraining for multilingual and robust speech recognition, yielding lower WERs than Large in most languages.

---

## Intended use

### Primary use cases

* High-accuracy Spanish speech transcription  
* Long-form audio content like podcasts, interviews, or lectures  
* ASR research and benchmarking in Spanish  

### Limitations

* Performance may degrade on highly noisy audio, heavy regional accents, or spontaneous speech  
* High computational cost for real-time inference  
* Not validated for legal, medical, or safety-critical transcription without human review  

---

## Training and evaluation data

* **Dataset:** Mozilla Common Voice 13.0 (Spanish subset)  
* **Data type:** Crowd-sourced read speech  
* **Preprocessing:**  
  * Audio resampled to 16 kHz  
  * Text tokenized using Whisper tokenizer  
  * Removal of invalid or corrupted samples  

* **Evaluation metric:** Word Error Rate (WER) on held-out evaluation set  

---

## Evaluation results

| Metric     | Value      |
| ---------- | ---------- |
| WER (eval) | **4.8949%** |

---

## Training procedure

### Training hyperparameters

* Learning rate: 1e-5  
* Optimizer: Adam (β1=0.9, β2=0.999, ε=1e-8)  
* LR scheduler: Linear  
* Warmup steps: 500  
* Training steps: 20000  
* Train batch size: 32 (gradient accumulation 2 → effective batch size 64)  
* Eval batch size: 16  
* Seed: 42  

### Training results (summary)

| Training Loss | Epoch | Step  | Validation Loss | WER    |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 0.0869        | 2.0   | 1000  | 0.1754          | 6.1516 |
| 0.0913        | 4.0   | 2000  | 0.1652          | 5.7500 |
| 0.051         | 6.0   | 3000  | 0.1643          | 5.7757 |
| 0.0391        | 8.0   | 4000  | 0.1881          | 5.6589 |
| 0.0104        | 10.0  | 5000  | 0.2026          | 5.6211 |
| 0.0806        | 12.01 | 6000  | 0.1741          | 5.7398 |
| 0.0077        | 14.01 | 7000  | 0.2119          | 5.6038 |
| 0.0357        | 16.01 | 8000  | 0.1776          | 5.6147 |
| 0.1087        | 18.01 | 9000  | 0.1868          | 5.5172 |
| 0.0401        | 20.01 | 10000 | 0.2014          | 5.4428 |
| 0.0334        | 22.01 | 11000 | 0.1751          | 5.2824 |
| 0.0071        | 24.01 | 12000 | 0.2295          | 5.2490 |
| 0.0374        | 26.01 | 13000 | 0.2098          | 5.2574 |
| 0.0023        | 28.01 | 14000 | 0.2498          | 5.0418 |
| 0.0025        | 30.01 | 15000 | 0.2311          | 4.9385 |
| 0.0006        | 32.01 | 16000 | 0.2544          | 4.8949 |
| 0.0009        | 34.02 | 17000 | 0.2691          | 5.1246 |
| 0.003         | 36.02 | 18000 | 0.2249          | 5.0277 |
| 0.0009        | 38.02 | 19000 | 0.2603          | 5.0373 |
| 0.0008        | 40.02 | 20000 | 0.2657          | 5.0225 |

---

## Framework versions

- Transformers 4.33.0.dev0  
- PyTorch 2.0.1+cu117  
- Datasets 2.14.4  
- Tokenizers 0.13.3  

---

## Example usage

```python
from transformers import pipeline

hf_model = "HiTZ/whisper-large-v2-es"  # replace with actual repo ID
device = 0  # -1 for CPU

pipe = pipeline(
    task="automatic-speech-recognition",
    model=hf_model,
    device=device
)

result = pipe("audio.wav")
print(result["text"])
```

---

## Ethical considerations and risks

* This model transcribes speech and may process personal data.
* Users should ensure compliance with applicable data protection laws (e.g., GDPR).
* The model should not be used for surveillance or non-consensual audio processing.

---

## Citation

If you use this model in your research, please cite:

```bibtex
@misc{dezuazo2025whisperlmimprovingasrmodels,
  title={Whisper-LM: Improving ASR Models with Language Models for Low-Resource Languages},
  author={Xabier de Zuazo and Eva Navas and Ibon Saratxaga and Inma Hernáez Rioja},
  year={2025},
  eprint={2503.23542},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}
```

Please, check the related paper preprint in
[arXiv:2503.23542](https://arxiv.org/abs/2503.23542)
for more details.

---

## License

This model is available under the
[Apache-2.0 License](https://www.apache.org/licenses/LICENSE-2.0).
You are free to use, modify, and distribute this model as long as you credit
the original creators.

---

## Contact and attribution

* Fine-tuning and evaluation: HiTZ/Aholab - Basque Center for Language Technology
* Base model: OpenAI Whisper
* Dataset: Mozilla Common Voice

For questions or issues, please open an issue in the model repository.