<!-- 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-base-timit-demo-colab
This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4779
- Wer: 0.3468
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: 0.0001
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
3.4408 | 4.0 | 500 | 1.2302 | 0.9116 |
0.561 | 8.0 | 1000 | 0.4809 | 0.4320 |
0.2091 | 12.0 | 1500 | 0.4285 | 0.3880 |
0.1221 | 16.0 | 2000 | 0.4448 | 0.3665 |
0.0858 | 20.0 | 2500 | 0.4622 | 0.3585 |
0.0597 | 24.0 | 3000 | 0.4621 | 0.3517 |
0.0453 | 28.0 | 3500 | 0.4779 | 0.3468 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.10.3