wav2vec2-1

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: 0.5980
• Wer: 0.4949

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: 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: 400
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.2691	1.37	200	2.9045	1.0
1.6356	2.74	400	0.9277	0.8678
0.8062	4.11	600	0.8200	0.7776
0.5983	5.48	800	0.6829	0.7161
0.4863	6.85	1000	0.6205	0.6507
0.407	8.22	1200	0.6519	0.6763
0.3641	9.59	1400	0.5771	0.6088
0.3291	10.96	1600	0.6548	0.6202
0.2905	12.33	1800	0.6538	0.5828
0.2613	13.7	2000	0.6281	0.5864
0.2354	15.07	2200	0.5936	0.5630
0.2145	16.44	2400	0.5877	0.5699
0.2008	17.81	2600	0.5469	0.5488
0.1751	19.18	2800	0.6453	0.5584
0.169	20.55	3000	0.5871	0.5357
0.1521	21.92	3200	0.6063	0.5318
0.1426	23.29	3400	0.5609	0.5171
0.1287	24.66	3600	0.6056	0.5126
0.1236	26.03	3800	0.5994	0.5074
0.1138	27.4	4000	0.5980	0.4944
0.1083	28.77	4200	0.5980	0.4949

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1