Wav2Vec2 Adult/Child Indonesian Speech Classifier
Wav2Vec2 Adult/Child Indonesian Speech Classifier is an audio classification model based on the wav2vec 2.0 architecture. This model is a fine-tuned version of wav2vec2-base on a private adult/child Indonesian speech classification dataset.
This model was trained using HuggingFace's PyTorch framework. All training was done on a Tesla P100, provided by Kaggle. Training metrics were logged via Tensorboard.
Model
Model | #params | Arch. | Training/Validation data (text) |
---|---|---|---|
wav2vec2-adult-child-id-cls |
91M | wav2vec 2.0 | Adult/Child Indonesian Speech Classification Dataset |
Evaluation Results
The model achieves the following results on evaluation:
Dataset | Loss | Accuracy | F1 |
---|---|---|---|
Adult/Child Indonesian Speech Classification | 0.2603 | 92.22% | 0.9202 |
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
learning_rate
: 3e-05train_batch_size
: 32eval_batch_size
: 32seed
: 42optimizer
: Adam withbetas=(0.9,0.999)
andepsilon=1e-08
lr_scheduler_type
: linearlr_scheduler_warmup_ratio
: 0.1gradient_accumulation_steps
: 1num_epochs
: 5
Training results
Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 |
---|---|---|---|---|---|
0.2415 | 1.0 | 305 | 0.2951 | 0.8804 | 0.8695 |
0.202 | 2.0 | 610 | 0.2392 | 0.9124 | 0.9081 |
0.2161 | 3.0 | 915 | 0.2508 | 0.9199 | 0.9161 |
0.1348 | 4.0 | 1220 | 0.2748 | 0.9153 | 0.9126 |
0.162 | 5.0 | 1525 | 0.2603 | 0.9222 | 0.9202 |
Disclaimer
Do consider the biases which came from pre-training datasets that may be carried over into the results of this model.
Authors
Wav2Vec2 Adult/Child Indonesian Speech Classifier was trained and evaluated by Ananto Joyoadikusumo. All computation and development are done on Kaggle.
Framework versions
- Transformers 4.18.0
- Pytorch 1.11.0+cu102
- Datasets 2.2.0
- Tokenizers 0.12.1
- Downloads last month
- 6