pic_asr-scr_w2v2-base_001

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: 0.9212
• Per: 0.1186
• Pcc: 0.6654
• Ctc Loss: 0.4168
• Mse Loss: 0.8851

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: 16
• eval_batch_size: 1
• seed: 1111
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 749
• training_steps: 7490

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
12.2064	1.0	749	4.4327	0.9994	0.6002	3.7601	0.8746
3.5366	2.0	1498	2.1284	0.2434	0.6278	0.9770	1.1415
1.435	3.0	2247	1.6059	0.1732	0.6394	0.5566	0.9882
1.021	4.0	2996	1.5984	0.1413	0.6353	0.4916	1.0429
0.637	5.0	3745	1.3181	0.1324	0.6561	0.4527	0.8927
0.2645	6.0	4494	1.5583	0.1253	0.6551	0.4269	1.1025
-0.1157	7.0	5243	1.2774	0.1240	0.6638	0.4200	0.9795
-0.4455	8.0	5992	1.3101	0.1205	0.6682	0.4151	1.0410
-0.7058	9.0	6741	0.9801	0.1188	0.6643	0.4166	0.9019
-0.8607	10.0	7490	0.9212	0.1186	0.6654	0.4168	0.8851

Framework versions

• Transformers 4.38.1
• Pytorch 2.0.1
• Datasets 2.16.1
• Tokenizers 0.15.2