Whisper Base Spanish is an automatic speech recognition (ASR) model for Spanish (es) speech. It is fine-tuned from [openai/whisper-base] on the Spanish subset of Mozilla Common Voice 13.0, achieving a Word Error Rate (WER) of 13.5312% on the evaluation split.

This variant balances transcription quality and model size, suitable for general-purpose Spanish ASR tasks.

Model description

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

Fine-tuned for improved transcription quality over Whisper Tiny while remaining lightweight compared to larger Whisper models.

Intended use

Primary use cases

General-purpose Spanish speech transcription
Research and experimentation with Spanish ASR
Moderate resource environments where Whisper Large is too heavy

Out-of-scope use

Professional transcription requiring near-zero WER
Very noisy or heavily accented Spanish
Safety-critical applications

Limitations and known issues

Performance may vary on:
- Noisy audio or multi-speaker recordings
- Regional dialects not well represented in Common Voice
- Rapid conversational speech
While better than Whisper Tiny, it may still produce transcription errors in difficult conditions.

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 normalized using Whisper tokenizer
- Invalid samples removed
Evaluation metric: Word Error Rate (WER) on held-out evaluation set

Evaluation results

Metric	Value
WER (eval)	13.5312%

Training procedure

Training hyperparameters

Learning rate: 2.5e-5
Optimizer: Adam (β1=0.9, β2=0.999, ε=1e-8)
LR scheduler: Linear
Warmup steps: 500
Training steps: 5000
Train batch size: 128
Eval batch size: 64
Seed: 42

Training results (summary)

Training Loss	Epoch	Step	Validation Loss	WER
0.2173	4.0	1000	0.3409	14.8123
0.0955	8.01	2000	0.3377	15.4269
0.1647	12.01	3000	0.3393	14.5602
0.0986	16.01	4000	0.3281	13.5312
0.1272	20.02	5000	0.3423	13.7596

Framework versions

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

How to use

python

from transformers import pipeline

hf_model = "HiTZ/whisper-base-es"  # 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 for more details.

License

This model is available under the Apache-2.0 License. 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.