dat259-cv_en-wav2vec2

This model is a fine-tuned version of facebook/wav2vec2-base on the common_voice_1_0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.4286
• Wer: 0.5339

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.383	1.32	200	3.0312	1.0
2.7346	2.63	400	2.0207	0.9413
0.8772	3.95	600	1.3528	0.6911
0.4718	5.26	800	1.1830	0.6135
0.3209	6.58	1000	1.2242	0.5856
0.2556	7.89	1200	1.3165	0.5894
0.2055	9.21	1400	1.3716	0.5708
0.18	10.53	1600	1.2709	0.5702
0.1497	11.84	1800	1.4300	0.5526
0.126	13.16	2000	1.3852	0.5507
0.1145	14.47	2200	1.3893	0.5489
0.101	15.79	2400	1.4051	0.5431
0.091	17.11	2600	1.4103	0.5399
0.0793	18.42	2800	1.3968	0.5324
0.0746	19.74	3000	1.4286	0.5339

Framework versions

• Transformers 4.21.1
• Pytorch 1.12.1+cu102
• Datasets 2.4.0
• Tokenizers 0.12.1