osman31/whisper-tiny-urdu-v1

whisper-tiny-urdu-v2

This model is a fine-tuned version of openai/whisper-tiny on the Mozilla Common Voice dataset v22 for Urdu. It achieves the following results on the evaluation set:

• Loss: 0.7495
• Wer: 51.2522

This fine-tuning resulted in bringing down the Word Error Rate (WER) from 125% to 51%!

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: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 250
• training_steps: 3000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7431	1.1088	500	0.8535	57.2363
0.5401	2.2175	1000	0.7817	53.9931
0.4691	3.3263	1500	0.7566	53.5244
0.4188	4.4351	2000	0.7513	52.5423
0.3595	5.5438	2500	0.7486	52.3570
0.3441	6.6526	3000	0.7495	51.2522

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.4.0
• Tokenizers 0.21.0