wav2vec2-xlsr-vorarlbergerisch

This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-xlsr-53-german on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.3193
• Wer: 0.3235

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
15.6717	3.83	100	3.0247	1.0
2.485	7.68	200	1.5937	0.9046
0.784	11.53	300	1.2664	0.5
0.3689	15.38	400	1.2046	0.4696
0.2618	19.23	500	1.1289	0.4155
0.2088	23.08	600	0.9339	0.3623
0.1388	26.91	700	1.1448	0.3573
0.1042	30.75	800	1.1411	0.3606
0.0784	34.6	900	1.2046	0.3547
0.0607	38.45	1000	1.2243	0.3488
0.0459	42.3	1100	1.2387	0.3226
0.0273	46.15	1200	1.2123	0.3387
0.0195	49.98	1300	1.2232	0.3345
0.0188	53.83	1400	1.2656	0.3235
0.0132	57.68	1500	1.3377	0.3285
0.0089	61.53	1600	1.3193	0.3235

Framework versions

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