wav2vec2-Vocals-Kor

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4436
• Cer: 0.2135

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: 16
• 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: 12
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
10.0761	0.1181	300	3.5931	0.9861
2.8824	0.2361	600	1.9956	0.6491
1.1701	0.3542	900	0.8263	0.2735
0.8015	0.4723	1200	0.6946	0.2530
0.7235	0.5903	1500	0.6638	0.2380
0.6747	0.7084	1800	0.6288	0.2399
0.6528	0.8264	2100	0.5963	0.2382
0.6185	0.9445	2400	0.6014	0.2412
0.5861	1.0626	2700	0.5747	0.2388
0.5668	1.1806	3000	0.5561	0.2199
0.5628	1.2987	3300	0.5335	0.2235
0.5521	1.4168	3600	0.5489	0.2290
0.5309	1.5348	3900	0.4995	0.2125
0.5033	1.6529	4200	0.4905	0.2171
0.5018	1.7710	4500	0.4853	0.2129
0.5011	1.8890	4800	0.4901	0.2171
0.4907	2.0071	5100	0.4828	0.2135
0.4578	2.1251	5400	0.4855	0.2180
0.4552	2.2432	5700	0.4621	0.2216
0.4345	2.3613	6000	0.4669	0.2152
0.4332	2.4793	6300	0.4639	0.2171
0.4338	2.5974	6600	0.4517	0.2180
0.4181	2.7155	6900	0.4407	0.2117
0.4048	2.8335	7200	0.4394	0.2063
0.4003	2.9516	7500	0.4478	0.2100
0.3847	3.0697	7800	0.4478	0.2159
0.3634	3.1877	8100	0.4378	0.2145
0.3629	3.3058	8400	0.4386	0.2060
0.3603	3.4238	8700	0.4411	0.2127
0.361	3.5419	9000	0.4436	0.2135

Framework versions

• Transformers 4.40.1
• Pytorch 2.0.1+cu117
• Datasets 2.19.0
• Tokenizers 0.19.1