HiTZ/whisper-large-v3-eu

Whisper Large-V3 Basque

This model is a fine-tuned version of openai/whisper-large-v3 on the mozilla-foundation/common_voice_13_0 eu dataset. It achieves the following results on the evaluation set:

• Loss: 0.3803
• Wer: 10.6201

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0326	4.85	1000	0.2300	13.3278
0.004	9.71	2000	0.2723	12.2038
0.0058	14.56	3000	0.2771	12.4246
0.003	19.42	4000	0.2838	12.2119
0.003	24.27	5000	0.2740	11.7704
0.0014	29.13	6000	0.2936	11.5436
0.0015	33.98	7000	0.2911	11.5193
0.0012	38.83	8000	0.2939	11.3674
0.0009	43.69	9000	0.3039	11.4140
0.0002	48.54	10000	0.3063	10.9624
0.0009	53.4	11000	0.3014	11.3350
0.0011	58.25	12000	0.3052	11.0474
0.0001	63.11	13000	0.3204	10.8692
0.0	67.96	14000	0.3413	10.7092
0.0	72.82	15000	0.3524	10.6647
0.0	77.67	16000	0.3607	10.6566
0.0	82.52	17000	0.3675	10.6120
0.0	87.38	18000	0.3737	10.6140
0.0	92.23	19000	0.3782	10.6181
0.0	97.09	20000	0.3803	10.6201

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1

Citation

If you use these models in your research, please cite:

@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},
      url={https://arxiv.org/abs/2503.23542}, 
}

Please, check the related paper preprint in arXiv:2503.23542 for more details.

Licensing

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.