wav2vec2-1b-Elderly4

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3586
• Cer: 9.6863

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: 0.0001
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• 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: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
8.3662	0.2580	200	1.7987	35.5146
1.448	0.5160	400	1.1509	26.5801
1.1188	0.7741	600	1.0576	24.9471
0.9565	1.0321	800	0.9458	23.6783
0.7732	1.2901	1000	0.7574	18.7911
0.7109	1.5481	1200	0.6910	17.4636
0.6921	1.8062	1400	0.6866	18.4387
0.6118	2.0642	1600	0.6038	15.8071
0.5056	2.3222	1800	0.6301	16.7058
0.4852	2.5802	2000	0.5687	14.9495
0.4708	2.8383	2200	0.5481	15.1257
0.4406	3.0963	2400	0.4806	13.2049
0.3592	3.3543	2600	0.4766	12.5529
0.3367	3.6123	2800	0.4146	11.3076
0.3173	3.8703	3000	0.3976	10.9492
0.274	4.1284	3200	0.4133	10.8964
0.2338	4.3864	3400	0.3755	10.1797
0.227	4.6444	3600	0.3729	9.8978
0.2144	4.9024	3800	0.3586	9.6863

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.1.post100
• Datasets 2.19.1
• Tokenizers 0.20.1