GerwinVanGiessen/whisper-base-nl-1

Whisper Base NL

This model is a fine-tuned version of openai/whisper-base on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.343928
• Wer: 19.003155

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
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 7500
• mixed_precision_training: Native AMP

Training results

Training Loss	Step	Validation Loss	Wer
0.3639	500	0.396971	24.3028
0.2625	1000	0.358340	22.5210
0.2212	1500	0.341232	21.0322
0.1455	2000	0.330033	20.2046
0.1406	2500	0.324484	20.0508
0.1244	3000	0.321562	19.5279
0.0848	3500	0.321506	19.5114
0.0844	4000	0.316492	19.1462
0.0731	4500	0.321992	19.0167
0.0515	5000	0.324720	19.1492
0.0532	5500	0.324773	19.0148
0.0426	6000	0.332404	19.0576
0.0328	6500	0.334900	18.8249
0.0327	7000	0.335876	19.0080
0.0252	7500	0.343928	19.0031

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1