models

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

• Loss: 0.0169
• Wer: 0.2673

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.6087	1.26	500	3.0092	1.0
2.1293	2.53	1000	0.4565	0.6872
0.443	3.79	1500	0.1887	0.3995
0.2368	5.05	2000	0.0564	0.3034
0.1775	6.31	2500	0.0398	0.2964
0.132	7.58	3000	0.0574	0.2895
0.1176	8.84	3500	0.0298	0.2749
0.1023	10.1	4000	0.0243	0.2708
0.0833	11.36	4500	0.0235	0.2796
0.0668	12.63	5000	0.0160	0.2714
0.0559	13.89	5500	0.0264	0.2749
0.0414	15.15	6000	0.0157	0.2673
0.0388	16.41	6500	0.0231	0.2772
0.0313	17.68	7000	0.0168	0.2667
0.0286	18.94	7500	0.0169	0.2673

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2