wav2vec2-base-cv

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

• Loss: 1.1562
• Wer: 0.3804

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
4.563	3.18	500	2.9826	1.0
2.0012	6.37	1000	0.9528	0.5354
0.4841	9.55	1500	0.8838	0.4325
0.2748	12.74	2000	0.9437	0.4130
0.1881	15.92	2500	0.9603	0.4005
0.1426	19.11	3000	1.0605	0.3955
0.1134	22.29	3500	1.0733	0.3897
0.0963	25.48	4000	1.1387	0.3835
0.0829	28.66	4500	1.1562	0.3804

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.13.3
• Tokenizers 0.10.3