Over the last few days, I have been fine-tuning Nvidia Parakeet for Basque, and I have published the result here.

The goal was simple: to have a lightweight Basque speech-to-text model that runs fast and is practical on modest hardware.

Accuracy vs speed

To be clear, this model is not as accurate as my best Basque Whisper model:

However, it has a major advantage: it is very fast and can run on CPU-only setups.

If maximum accuracy is your priority, basque Whisper Large v3 is still the better option. If you prioritize speed and low-resource inference, Parakeet is a strong alternative.

For practical usage, I recommend running this model in ONNX format.

Pre-converted ONNX files are available here:

This makes integration easier and gives good performance, especially in CPU environments.

About the fine-tuning process

More details about training, metrics, and tracking are available here:

Short conclusion

For Basque STT right now:

  • If you prioritize accuracy: whisper-large-v3-eu
  • If you prioritize speed and CPU usage: parakeet-tdt-0.6b-v3-basque (via ONNX)