wav2vec2-ctc

This model is a fine-tuned version of fav-kky/wav2vec2-base-cs-80k-ClTRUS on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.5716
• Wer: 0.9831

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• 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_ratio: 0.3
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.4027	2.1942	2000	3.5478	1.0
3.2864	4.3884	4000	3.4339	1.0
2.8505	6.5826	6000	2.8626	0.9998
2.5028	8.7767	8000	2.5475	0.9989
2.2853	10.9709	10000	2.3584	0.9989
2.1248	13.1651	12000	2.2147	0.9992
2.0013	15.3593	14000	2.1026	0.9989
1.8839	17.5535	16000	1.9910	0.9986
1.7885	19.7477	18000	1.9146	0.9991
1.7114	21.9419	20000	1.8460	0.9977
1.6532	24.1360	22000	1.7951	0.9968
1.606	26.3302	24000	1.7419	0.9959
1.5582	28.5244	26000	1.7155	0.9958
1.5257	30.7186	28000	1.6877	0.9917
1.4951	32.9128	30000	1.6671	0.9923
1.4748	35.1070	32000	1.6511	0.9902
1.4454	37.3012	34000	1.6320	0.9891
1.4241	39.4953	36000	1.6238	0.9884
1.4038	41.6895	38000	1.6056	0.9843
1.3902	43.8837	40000	1.5955	0.9839
1.3795	46.0779	42000	1.5908	0.9848
1.3708	48.2721	44000	1.5855	0.9833
1.3633	50.4663	46000	1.5863	0.9858
1.3564	52.6604	48000	1.5791	0.9830
1.3542	54.8546	50000	1.5714	0.9818
1.3379	57.0488	52000	1.5722	0.9824
1.3441	59.2430	54000	1.5716	0.9831

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0