wav2vec2-base-vios-v4

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: 0.3198
• Wer: 0.2169

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: 5e-05
• 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: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.8138	0.69	500	3.5011	1.0
3.4372	1.37	1000	3.3447	1.0
1.9519	2.06	1500	0.8356	0.5944
0.8581	2.74	2000	0.5280	0.4038
0.6405	3.43	2500	0.4410	0.3410
0.5417	4.12	3000	0.3990	0.3140
0.4804	4.8	3500	0.3804	0.2973
0.4384	5.49	4000	0.3644	0.2808
0.4162	6.17	4500	0.3542	0.2648
0.3941	6.86	5000	0.3436	0.2529
0.3733	7.54	5500	0.3355	0.2520
0.3564	8.23	6000	0.3294	0.2415
0.3412	8.92	6500	0.3311	0.2332
0.3266	9.6	7000	0.3217	0.2325
0.3226	10.29	7500	0.3317	0.2303
0.3115	10.97	8000	0.3226	0.2279
0.3094	11.66	8500	0.3157	0.2236
0.2967	12.35	9000	0.3109	0.2202
0.2995	13.03	9500	0.3129	0.2156
0.2895	13.72	10000	0.3195	0.2146
0.3089	14.4	10500	0.3198	0.2169

Framework versions

• Transformers 4.16.2
• Pytorch 1.11.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1