wav2vec2-base_toy_train_data_masked_audio_10ms

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: 1.2477
• Wer: 0.7145

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: 1000
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.1337	1.05	250	3.4081	0.9982
3.0792	2.1	500	3.2446	0.9982
2.0577	3.15	750	1.5839	0.9492
1.3639	4.2	1000	1.3279	0.8798
1.0814	5.25	1250	1.1629	0.8294
0.8722	6.3	1500	1.1305	0.8140
0.7602	7.35	1750	1.1241	0.7972
0.6982	8.4	2000	1.1429	0.7780
0.6494	9.45	2250	1.1047	0.7620
0.5924	10.5	2500	1.1756	0.7649
0.5385	11.55	2750	1.2230	0.7736
0.5026	12.6	3000	1.1783	0.7472
0.4973	13.65	3250	1.1613	0.7287
0.4726	14.7	3500	1.1923	0.7345
0.4521	15.75	3750	1.2153	0.7171
0.4552	16.81	4000	1.2485	0.7226
0.422	17.86	4250	1.2664	0.7240
0.3708	18.91	4500	1.2352	0.7148
0.3516	19.96	4750	1.2477	0.7145

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu102
• Datasets 2.0.0
• Tokenizers 0.11.6