<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
wav2vec2-large-voxrex-swedish-coscan-no-region
This model is a fine-tuned version of KBLab/wav2vec2-large-voxrex-swedish on the coscan-speech2 dataset. It achieves the following results on the evaluation set:
- Loss: 1.0151
- Accuracy: 0.8773
- F1: 0.8773
- Precision: 0.8773
- Recall: 0.8773
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: 3e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 | Precision | Recall |
|---|---|---|---|---|---|---|---|
| 0.1651 | 1.0 | 6468 | 0.5657 | 0.8650 | 0.8650 | 0.8650 | 0.8650 |
| 0.1217 | 2.0 | 12936 | 0.9411 | 0.8487 | 0.8487 | 0.8487 | 0.8487 |
| 0.0013 | 3.0 | 19404 | 0.9991 | 0.8617 | 0.8617 | 0.8617 | 0.8617 |
| 0.0652 | 4.0 | 25872 | 1.0151 | 0.8773 | 0.8773 | 0.8773 | 0.8773 |
| 0.0001 | 5.0 | 32340 | 1.1031 | 0.8700 | 0.8700 | 0.8700 | 0.8700 |
Classification report on Coscan Speech (test set)
precision recall f1-score support
Bergen og Ytre Vestland 0.65 0.97 0.78 1809
Hedmark og Oppland 0.12 0.06 0.08 2302
Nordland 0.97 0.47 0.63 2195
Oslo-området 0.78 0.42 0.55 6957
Sunnmøre 0.94 0.71 0.81 2636
Sør-Vestlandet 0.96 0.46 0.62 2860
Sørlandet 0.62 0.81 0.70 2490
Troms 0.67 1.00 0.80 2867
Trøndelag 0.52 0.94 0.67 2666
Voss og omland 0.70 0.71 0.71 2641
Ytre Oslofjord 0.20 0.49 0.29 1678
accuracy 0.62 31101
macro avg 0.65 0.64 0.60 31101
weighted avg 0.68 0.62 0.61 31101
Framework versions
- Transformers 4.22.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 2.4.1.dev0
- Tokenizers 0.12.1
