wav2vec2-base-vios-v2

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: 0.6056
• Wer: 0.2442

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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.8344	0.69	500	3.5012	1.0
3.4505	1.37	1000	3.4081	1.0
2.1426	2.06	1500	0.8761	0.6241
0.8801	2.74	2000	0.5476	0.4241
0.6453	3.43	2500	0.4384	0.3495
0.5449	4.12	3000	0.4055	0.3160
0.4862	4.8	3500	0.3815	0.3002
0.4435	5.49	4000	0.3525	0.2776
0.4205	6.17	4500	0.3660	0.2725
0.3974	6.86	5000	0.3386	0.2565
0.3758	7.54	5500	0.3492	0.2607
0.3595	8.23	6000	0.3391	0.2441
0.3438	8.92	6500	0.3255	0.2354
0.3308	9.6	7000	0.3379	0.2422
0.3265	10.29	7500	0.3375	0.2349
0.311	10.97	8000	0.3356	0.2306
0.3071	11.66	8500	0.3286	0.2249
0.2941	12.35	9000	0.3176	0.2211
0.296	13.03	9500	0.3268	0.2257
0.2852	13.72	10000	0.3265	0.2196
0.3102	14.4	10500	0.3390	0.2209
0.2974	15.09	11000	0.3493	0.2199
0.3433	15.78	11500	0.3687	0.2199
0.3526	16.46	12000	0.3698	0.2170
0.36	17.15	12500	0.4110	0.2227
0.4322	17.83	13000	0.4830	0.2290
0.4973	18.52	13500	0.5280	0.2356
0.5701	19.2	14000	0.5990	0.2370
0.6014	19.89	14500	0.6056	0.2442

Framework versions

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