wav2vec2-dataset-vios

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

• Loss: 0.5423
• Wer: 0.4075

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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.0963	1.1	400	1.1336	0.7374
0.6576	2.2	800	0.4716	0.3727
0.4099	3.3	1200	0.3907	0.3100
0.3332	4.4	1600	0.3735	0.2766
0.2976	5.49	2000	0.3932	0.2801
0.2645	6.59	2400	0.3628	0.2542
0.2395	7.69	2800	0.3702	0.2734
0.2208	8.79	3200	0.3667	0.2467
0.1974	9.89	3600	0.3688	0.2398
0.1772	10.99	4000	0.3819	0.2457
0.1695	12.09	4400	0.3840	0.2451
0.319	13.19	4800	0.6531	0.4084
0.7305	14.29	5200	0.9883	0.6348
0.5787	15.38	5600	0.5260	0.3063
0.8558	16.48	6000	1.2870	0.7692
1.155	17.58	6400	1.0568	0.6353
0.8393	18.68	6800	0.7360	0.4486
0.6094	19.78	7200	0.6072	0.4108
0.5346	20.88	7600	0.5749	0.4095
0.5073	21.98	8000	0.5588	0.4056
0.4859	23.08	8400	0.5475	0.4015
0.475	24.18	8800	0.5430	0.4011
0.4683	25.27	9200	0.5400	0.3990
0.4673	26.37	9600	0.5407	0.4011
0.4665	27.47	10000	0.5408	0.3992
0.4703	28.57	10400	0.5420	0.4070
0.4709	29.67	10800	0.5423	0.4075

Framework versions

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