whisper-large-v3-persian

This model is a fine-tuned version of openai/whisper-large-v3 on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2499
• Wer: 26.5381

Model description

The data was fine-tuned using an RTX 6000 ADA graphics card. Over 200,000 samples were fine-tuned on the system. This data belonged to the Mozilla Foundation's Common Voice 17.0 dataset. The obtained result, despite improving the Word Error Rate (WER) compared to other models, still has grammatical weaknesses, which is due to spelling errors in the dataset.

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: 8
• seed: 42
• 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: 500
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1337	0.8110	2000	0.2818	31.0620
0.0608	1.6221	4000	0.2532	28.8171
0.0229	2.4331	6000	0.2499	26.5381

Framework versions

• Transformers 4.52.3
• Pytorch 2.7.0+cu126
• Datasets 3.6.0
• Tokenizers 0.21.1