Neprox/STT-swedish-extended-dataset-model

Whisper Small - Swedish

This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 & NST dataset. It achieves the following results on the evaluation set:

• Loss: 0.3551
• Wer: 19.2143

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2128	0.85	1000	0.2955	22.1613
0.0871	1.71	2000	0.2790	20.8034
0.0373	2.56	3000	0.2884	19.9269
0.0163	3.41	4000	0.3082	19.5477
0.0046	4.27	5000	0.3183	19.5881
0.0023	5.12	6000	0.3397	19.3757
0.0023	5.97	7000	0.3468	19.3219
0.0013	6.83	8000	0.3551	19.2143

Framework versions

• Transformers 4.25.0.dev0
• Pytorch 1.12.1
• Datasets 2.7.1
• Tokenizers 0.13.2