---
language: ["pt"]
tags:
- automatic-speech-recognition
- whisperx
- audio
- speech-to-text
- portuguese
license: mit
library_name: whisperx
base_model: openai/whisper-large-v3
datasets:
- your_dataset_name
metrics:
- wer
- cer
---
# WhisperX Large-v3 — PRE-ACORDO 

This model is a **fine-tuned [Whisper](https://github.com/m-bain/whisperX)** variant of `openai/whisper-large-v3`, trained for **European Portuguese (EP) (around 425h)** automatic speech recognition (ASR).
From [CAMÕES](https://arxiv.org/pdf/2508.19721) work. This is the best model for long-form inference. The text is normalized. 
---
## 🧠 Model Description
- **Base model:** `openai/whisper-large-v3`
- **Architecture:** Transformer encoder–decoder
- **Training:** Fine-tuned on around 425 hours of Portuguese speech
- **Task:** Transcription (`task="transcribe"`)
- **Compute type:** float16 (recommended)

## 🧩 Usage

```python
import whisperx

device = "cuda"  # or "cpu"
compute_type = "float16"

model = whisperx.load_model(
    "inesc-id/WhisperLv3-EP-X",
    device=device,
    compute_type=compute_type,
    language="pt",
    task="transcribe"
)

## Citation


**BibTeX:**

```
@inproceedings{camoes,
    title={{CAMÕES: A Comprehensive Automatic Speech Recognition Benchmark for  European Portuguese}},
    author={Carlos Carvalho, Francisco Teixeira, Catarina Botelho, Anna Pompili, Rubén Solera-Ureña, Sérgio Paulo, Mariana Julião, Thomas Rolland, John Mendonça, Diogo Pereira, Isabel Trancoso, Alberto Abad},
    booktitle={Proceedings of the IEEE Automatic Speech Recognition and Understanding Workshop (ASRU)},
    year={2025},
}
```