wav2vec2-vivos

This model is a fine-tuned version of facebook/wav2vec2-base on the vivos dataset. It achieves the following results on the evaluation set:

• Loss: 0.3014
• Wer: 0.1730

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: 0.0002
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.3
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.9735	2.0	146	3.8105	1.0
3.4566	4.0	292	3.5691	1.0
3.0957	6.0	438	1.5923	0.9375
0.7156	8.0	584	0.4616	0.3431
0.3195	10.0	730	0.3547	0.2452
0.2113	12.0	876	0.3117	0.2053
0.163	14.0	1022	0.3106	0.1868
0.1486	16.0	1168	0.3037	0.1748
0.1047	18.0	1314	0.3008	0.1728
0.0963	20.0	1460	0.3014	0.1730

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0
• Datasets 2.21.0
• Tokenizers 0.19.1