metadata

license: apache-2.0
tags:
  - generated_from_trainer
datasets:
  - dataset/riksdagen
metrics:
  - wer
model-index:
  - name: whisper-small-sv
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: dataset/riksdagen audiofolder
          type: dataset/riksdagen
          config: test
          split: test
          args: audiofolder
        metrics:
          - name: Wer
            type: wer
            value: 0.2426515530366172

whisper-small-sv

This model is a fine-tuned version of openai/whisper-small on the dataset/riksdagen audiofolder dataset. It achieves the following results on the evaluation set:

Loss: 0.3479
Wer: 0.2427

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: 64
eval_batch_size: 64
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 100
training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5024	0.04	250	0.5073	0.2948
0.4684	0.08	500	0.4639	0.2784
0.4246	0.12	750	0.4396	0.2758
0.4132	0.17	1000	0.4222	0.2664
0.4021	0.21	1250	0.4101	0.2633
0.3871	0.25	1500	0.3982	0.2619
0.3813	0.29	1750	0.3895	0.2577
0.3878	0.33	2000	0.3827	0.2533
0.3704	0.37	2250	0.3770	0.2533
0.3516	0.42	2500	0.3714	0.2540
0.3792	0.46	2750	0.3675	0.2495
0.3476	0.5	3000	0.3636	0.2456
0.3522	0.54	3250	0.3611	0.2462
0.3545	0.58	3500	0.3560	0.2440
0.3426	0.62	3750	0.3543	0.2464
0.3437	0.66	4000	0.3524	0.2464
0.3562	0.71	4250	0.3507	0.2452
0.3555	0.75	4500	0.3491	0.2426
0.3397	0.79	4750	0.3483	0.2419
0.3516	0.83	5000	0.3479	0.2427

Framework versions

Transformers 4.26.0.dev0
Pytorch 1.12.0a0+8a1a93a
Datasets 2.7.1
Tokenizers 0.13.2