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: 0.4229
• Wer: 0.2386

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
3.5486	4.0	500	2.1672	0.9876
0.6819	8.0	1000	0.4502	0.3301
0.2353	12.0	1500	0.4352	0.2841
0.1427	16.0	2000	0.4237	0.2584
0.0945	20.0	2500	0.4409	0.2545
0.0671	24.0	3000	0.4257	0.2413
0.0492	28.0	3500	0.4229	0.2386

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.18.3
• Tokenizers 0.10.3