wavlm-libri-clean-100h-base-plus
This model is a fine-tuned version of microsoft/wavlm-base-plus on the LIBRISPEECH_ASR - CLEAN dataset. It achieves the following results on the evaluation set:
- Loss: 0.0819
- Wer: 0.0683
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: 0.0003
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 32
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 3.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 2.8877 | 0.34 | 300 | 2.8649 | 1.0 |
| 0.2852 | 0.67 | 600 | 0.2196 | 0.1830 |
| 0.1198 | 1.01 | 900 | 0.1438 | 0.1273 |
| 0.0906 | 1.35 | 1200 | 0.1145 | 0.1035 |
| 0.0729 | 1.68 | 1500 | 0.1055 | 0.0955 |
| 0.0605 | 2.02 | 1800 | 0.0936 | 0.0859 |
| 0.0402 | 2.35 | 2100 | 0.0885 | 0.0746 |
| 0.0421 | 2.69 | 2400 | 0.0848 | 0.0700 |
Framework versions
- Transformers 4.15.0.dev0
- Pytorch 1.9.0+cu111
- Datasets 1.16.2.dev0
- Tokenizers 0.10.3