README.md

---
language:
- gl
license: apache-2.0
base_model: openai/whisper-large-v3
tags:
- whisper-event
- generated_from_trainer
datasets:
- mozilla-foundation/common_voice_13_0
metrics:
- wer
model-index:
- name: Whisper Large-V3 Galician
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: mozilla-foundation/common_voice_13_0 gl
      type: mozilla-foundation/common_voice_13_0
      config: gl
      split: test
      args: gl
    metrics:
    - name: Wer
      type: wer
      value: 5.008278145695364
---

# Whisper Large-V3 Galician

## Model summary

**Whisper Large-V3 Galician** is an automatic speech recognition (ASR) model for **Galician (gl)** speech. It is fine-tuned from [openai/whisper-large-v3] on the **Galician portion of Mozilla Common Voice 13.0**, achieving a **Word Error Rate (WER) of 5.01%** on the evaluation split.

This model is intended for high-accuracy transcription of Galician audio in research, media, and accessibility applications.

---

## Model description

* **Architecture:** Transformer-based encoder–decoder (Whisper)  
* **Base model:** openai/whisper-large-v3  
* **Language:** Galician (gl)  
* **Task:** Automatic Speech Recognition (ASR)  
* **Output:** Text transcription in Galician  
* **Decoding:** Autoregressive sequence-to-sequence decoding  

Fine-tuned on Galician speech data, leveraging Whisper’s multilingual pretraining for low-resource language transcription.

---

## Intended use

### Primary use cases

* High-quality transcription of Galician audio recordings  
* Offline or batch ASR pipelines  
* Research and development in Galician ASR  
* Media, educational, and accessibility tasks  

### Intended users

* Researchers working on Galician or low-resource ASR  
* Developers building Galician speech applications  
* Academic and institutional users  

### Out-of-scope use

* Real-time or low-latency ASR without optimization  
* Speech translation tasks  
* Safety-critical applications without additional validation  

---

## Limitations and known issues

* Performance may degrade on:
  * Noisy or low-quality recordings  
  * Conversational or spontaneous speech  
  * Accents underrepresented in Common Voice  
* Dataset biases may be reflected in outputs  
* Occasional transcription errors can occur under difficult acoustic conditions  

---

## Training and evaluation data

### Training data

* **Dataset:** Mozilla Common Voice 13.0 (Galician subset)  
* **Data type:** Crowd-sourced, read speech  
* **Preprocessing:**  
  * Audio resampled to 16 kHz  
  * Text normalized using Whisper tokenizer  
  * Filtering of invalid or problematic samples  

### Evaluation data

* **Dataset:** Mozilla Common Voice 13.0 (Galician evaluation split)  
* **Metric:** Word Error Rate (WER)  

---

## Evaluation results

| Metric     | Value      |
| ---------- | ---------- |
| WER (eval) | **5.01%** |

---

## 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: 20,000  
* Train batch size: 32  
* Eval batch size: 16  
* Gradient accumulation steps: 2  
* Total train batch size: 64  
* Seed: 42  
* Mixed precision: Native AMP  

### Training results (summary)

| Training Loss | Epoch | Step  | Validation Loss | WER    |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 0.0176        | 5.0   | 1000  | 0.1563          | 5.2514 |
| 0.004         | 10.0  | 2000  | 0.1884          | 5.5653 |
| 0.0039        | 15.0  | 3000  | 0.2052          | 5.5377 |
| 0.0033        | 20.0  | 4000  | 0.2054          | 5.2997 |
| 0.0012        | 25.0  | 5000  | 0.2115          | 5.1031 |
| 0.001         | 30.0  | 6000  | 0.2195          | 5.2394 |
| 0.001         | 35.0  | 7000  | 0.2257          | 5.3446 |
| 0.001         | 40.0  | 8000  | 0.2178          | 5.4015 |
| 0.0008        | 45.0  | 9000  | 0.2250          | 5.4705 |
| 0.0008        | 50.0  | 10000 | 0.2320          | 5.2946 |
| 0.0002        | 55.0  | 11000 | 0.2368          | 5.3515 |
| 0.0           | 60.0  | 12000 | 0.2551          | 5.0997 |
| 0.0           | 65.0  | 13000 | 0.2634          | 5.0738 |
| 0.0           | 70.0  | 14000 | 0.2697          | 5.0359 |
| 0.0           | 75.0  | 15000 | 0.2752          | 5.0186 |
| 0.0           | 80.0  | 16000 | 0.2804          | 5.0066 |
| 0.0           | 85.0  | 17000 | 0.2852          | 4.9859 |
| 0.0           | 90.0  | 18000 | 0.2894          | 4.9893 |
| 0.0           | 95.0  | 19000 | 0.2927          | 5.0014 |
| 0.0           | 100.0 | 20000 | 0.2940          | 5.0083 |

---

## Framework versions

- Transformers 4.37.2  
- PyTorch 2.2.0+cu121  
- Datasets 2.16.1  
- Tokenizers 0.15.1  

---

## How to use

```python
from transformers import pipeline

hf_model = "HiTZ/whisper-large-v3-gl"  # replace with actual repo ID
device = 0  # set to -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.