saq-20s_asr-scr_w2v2-base_003

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: 1.4770
• Per: 0.1582
• Pcc: 0.6742
• Ctc Loss: 0.5610
• Mse Loss: 0.8995

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 1
• seed: 3333
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2226
• training_steps: 22260
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
16.8423	3.0	2226	4.5928	0.9983	0.6376	3.7671	0.8900
4.329	6.0	4452	4.5763	0.9983	0.6939	3.7500	0.9944
3.9518	9.0	6678	4.3570	0.9983	0.6962	3.7043	0.9024
3.4509	12.0	8904	3.2020	0.8036	0.6831	2.6097	0.7935
1.6554	15.0	11130	1.7792	0.2361	0.6712	0.9476	0.8436
0.9205	18.0	13356	1.6586	0.1900	0.6858	0.7026	0.9314
0.7081	21.0	15582	1.5955	0.1721	0.6716	0.6236	0.9429
0.5965	24.0	17808	1.5239	0.1634	0.6784	0.5880	0.9140
0.5351	27.0	20034	1.4992	0.1595	0.6782	0.5682	0.9115
0.5012	30.0	22260	1.4770	0.1582	0.6742	0.5610	0.8995

Framework versions

• Transformers 4.38.1
• Pytorch 2.0.1
• Datasets 2.18.0
• Tokenizers 0.15.2