wav2vec2-vorarlbergerisch

This model is a fine-tuned version of facebook/wav2vec2-base-960h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.9241
• Wer: 0.4358

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 62
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
12.6837	3.83	100	3.7188	1.0
3.33	7.68	200	3.0620	1.0
2.9508	11.53	300	2.5915	1.0101
1.8954	15.38	400	1.6930	0.8243
1.231	19.23	500	1.7179	0.7551
0.9862	23.08	600	1.5237	0.6529
0.7353	26.91	700	1.5119	0.5921
0.5368	30.75	800	1.5011	0.5574
0.4448	34.6	900	1.5334	0.5363
0.3278	38.45	1000	1.7125	0.5144
0.2575	42.3	1100	1.6529	0.4958
0.1966	46.15	1200	1.7670	0.4848
0.1552	49.98	1300	1.7586	0.4620
0.1118	53.83	1400	1.7912	0.4417
0.0847	57.68	1500	1.8709	0.4443
0.0654	61.53	1600	1.9241	0.4358

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu113
• Datasets 1.18.3
• Tokenizers 0.10.3