Modelpublic

slplab/wav2vec2-xlsr-korean-aihub-900

sourceHugging Faceapache-2.0updated 4y agoView on Hugging Face

0likes44downloads

Model Card

wav2vec2-xlsr-korean-aihub-900

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.2659
—Wer: 0.1089

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
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 700
—num_epochs: 50
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	PER
7.0213	3.51	200	2.5580	0.9872
2.4456	7.02	400	2.4060	0.9872
2.2322	10.53	600	2.0604	0.9872
1.9241	14.04	800	1.1675	0.5440
0.8402	17.54	1000	0.4376	0.2035
0.4796	21.05	1200	0.3261	0.1540
0.339	24.56	1400	0.2918	0.1375
0.2688	28.07	1600	0.2872	0.1310
0.2338	31.58	1800	0.2732	0.1218
0.212	35.09	2000	0.2654	0.1166
0.1939	38.6	2200	0.2658	0.1087
0.1738	42.11	2400	0.2692	0.1094
0.1601	45.61	2600	0.2666	0.1107
0.1587	49.12	2800	0.2659	0.1089

Framework versions

—Transformers 4.21.3
—Pytorch 1.12.1
—Datasets 2.4.0
—Tokenizers 0.12.1