xlsr-wav2vec2-1

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: 0.5437
• Wer: 0.4412

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: 0.0003
• 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: 800
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.517	1.38	400	3.0431	1.0
1.8387	2.76	800	0.6552	0.7263
0.5971	4.14	1200	0.5308	0.5885
0.4153	5.52	1600	0.4667	0.5551
0.3388	6.9	2000	0.4428	0.5260
0.2803	8.28	2400	0.4915	0.5164
0.2613	9.65	2800	0.4904	0.4988
0.237	11.03	3200	0.4998	0.5075
0.2175	12.41	3600	0.4905	0.4983
0.1969	13.79	4000	0.4818	0.4877
0.1932	15.17	4400	0.5578	0.5006
0.1782	16.55	4800	0.4981	0.4949
0.1655	17.93	5200	0.4978	0.4940
0.1505	19.31	5600	0.5360	0.4896
0.1362	20.69	6000	0.5441	0.4709
0.1246	22.07	6400	0.5358	0.4650
0.1117	23.45	6800	0.5513	0.4716
0.107	24.83	7200	0.5344	0.4578
0.0963	26.21	7600	0.5073	0.4452
0.0846	27.59	8000	0.5335	0.4497
0.0799	28.96	8400	0.5437	0.4412

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1