wav2vec2-balti-specaugment

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

Model description

More information needed

More information needed

More information needed

The following hyperparameters were used during training:

learning_rate: 0.0002
train_batch_size: 2
eval_batch_size: 2
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 300
num_epochs: 15
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
25.6751	0.3971	200	3.1876	1.0
22.3218	0.7942	400	2.3792	0.9999
7.6844	1.1906	600	0.7101	0.5854
5.7257	1.5877	800	0.5493	0.4532
5.3473	1.9849	1000	0.4829	0.3935
4.6374	2.3812	1200	0.4503	0.3841
4.4538	2.7784	1400	0.4169	0.3531
3.6916	3.1747	1600	0.3878	0.3464
3.7018	3.5719	1800	0.3864	0.3206
3.4336	3.9690	2000	0.3622	0.3303
3.1114	4.3654	2200	0.3546	0.3084
3.2920	4.7625	2400	0.3416	0.2986
2.7648	5.1588	2600	0.3389	0.2966
2.8410	5.5560	2800	0.3220	0.2858
2.8452	5.9531	3000	0.3281	0.2724
2.4203	6.3495	3200	0.3261	0.2707
2.6664	6.7466	3400	0.3093	0.2632
2.2534	7.1430	3600	0.3200	0.2630
2.3473	7.5401	3800	0.3253	0.2650
2.3196	7.9372	4000	0.3189	0.2541
2.0804	8.3336	4200	0.3056	0.2548
1.9665	8.7307	4400	0.3040	0.2529
1.7436	9.1271	4600	0.3207	0.2458
1.8735	9.5242	4800	0.3080	0.2428
1.7719	9.9213	5000	0.3086	0.2378
1.6804	10.3177	5200	0.3063	0.2428
1.9388	10.7148	5400	0.2929	0.2402
1.4199	11.1112	5600	0.3140	0.2325
1.6097	11.5083	5800	0.3064	0.2308
1.4726	11.9054	6000	0.2944	0.2324
1.5519	12.3018	6200	0.2969	0.2251
1.1738	12.6989	6400	0.3051	0.2252
1.2782	13.0953	6600	0.2973	0.2243
1.2456	13.4924	6800	0.2943	0.2248
1.3731	13.8896	7000	0.2882	0.2227
1.4417	14.2859	7200	0.2832	0.2211
1.3168	14.6830	7400	0.2866	0.2212