wav2vec-xlsr-53-turkish-v4

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8656
• Wer: 0.5719
• Cer: 0.1447

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: 3e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 750
• num_epochs: 12
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.1113	1.3498	1000	3.0797	1.0	1.0
0.8723	2.6995	2000	0.5867	0.5040	0.1197
0.752	4.0486	3000	0.4962	0.4101	0.0981
0.7464	5.3984	4000	0.5006	0.4184	0.0998
0.7835	6.7481	5000	0.5443	0.4332	0.1036
1.0919	8.0972	6000	0.6643	0.4878	0.1186
1.3457	9.4470	7000	0.9190	0.6869	0.1907
1.2715	10.7968	8000	0.8656	0.5719	0.1447

Framework versions

• Transformers 4.56.1
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.0