metadata
license: apache-2.0
base_model: facebook/wav2vec2-base
tags:
- generated_from_trainer
metrics:
- accuracy
model-index:
- name: emotions_6_classes_small
results: []
emotions_6_classes_small
This model is a fine-tuned version of facebook/wav2vec2-base on the 'Audio emotions' public dataset, available form https://www.kaggle.com/datasets/uldisvalainis/audio-emotions. 'Surprised' class was discarded due to lack of samples.
It achieves the following results on the evaluation set:
- Loss: 0.9106
- Accuracy: 0.7920
Model description
Classifies audios into 6 emotions:
- Angry
- Happy
- Sad
- Neutral
- Fearful
- Disgusted
Intended uses & limitations
This model was trained for educational purposes.
Training and evaluation data
- Training: 80%
- Test: 20%
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 3e-05
- train_batch_size: 128
- eval_batch_size: 128
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 512
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 30
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 0.2009 | 0.99 | 19 | 0.6892 | 0.7891 |
| 0.2272 | 1.97 | 38 | 0.7235 | 0.7817 |
| 0.2196 | 2.96 | 57 | 0.7027 | 0.7809 |
| 0.2402 | 4.0 | 77 | 0.7953 | 0.7592 |
| 0.2301 | 4.99 | 96 | 0.7979 | 0.7699 |
| 0.1896 | 5.97 | 115 | 0.7533 | 0.7838 |
| 0.188 | 6.96 | 134 | 0.7483 | 0.7817 |
| 0.1573 | 8.0 | 154 | 0.8200 | 0.7756 |
| 0.1576 | 8.99 | 173 | 0.7623 | 0.7944 |
| 0.1452 | 9.97 | 192 | 0.7460 | 0.7944 |
| 0.1322 | 10.96 | 211 | 0.8031 | 0.7875 |
| 0.1353 | 12.0 | 231 | 0.7864 | 0.7883 |
| 0.1211 | 12.99 | 250 | 0.7934 | 0.7903 |
| 0.1165 | 13.97 | 269 | 0.7734 | 0.7936 |
| 0.0928 | 14.96 | 288 | 0.8743 | 0.7842 |
| 0.095 | 16.0 | 308 | 0.8483 | 0.7867 |
| 0.0824 | 16.99 | 327 | 0.8860 | 0.7850 |
| 0.0896 | 17.97 | 346 | 0.8314 | 0.7957 |
| 0.0874 | 18.96 | 365 | 0.8164 | 0.7936 |
| 0.081 | 20.0 | 385 | 0.8250 | 0.7993 |
| 0.0673 | 20.99 | 404 | 0.9118 | 0.7879 |
| 0.0716 | 21.97 | 423 | 0.8605 | 0.7912 |
| 0.0588 | 22.96 | 442 | 0.8470 | 0.7985 |
| 0.0579 | 24.0 | 462 | 0.8906 | 0.7920 |
| 0.0511 | 24.99 | 481 | 0.8853 | 0.7969 |
| 0.0488 | 25.97 | 500 | 0.8901 | 0.7973 |
| 0.0468 | 26.96 | 519 | 0.9083 | 0.7895 |
| 0.0505 | 28.0 | 539 | 0.9010 | 0.7903 |
| 0.0542 | 28.99 | 558 | 0.8924 | 0.7944 |
| 0.0542 | 29.61 | 570 | 0.9106 | 0.7920 |
Framework versions
- Transformers 4.31.0
- Pytorch 2.0.1+cu117
- Datasets 2.14.4
- Tokenizers 0.13.3