sgangireddy/whisper-medium-tr

Whisper Medium Turkish

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_11_0 Turkish dataset. It achieves the following results on the evaluation set:

• Loss: 0.1879
• Wer: 10.5033

Model description

The model is fine-tuned for 1000 steps/updates. - Zero-shot - 20.89 (CV11) - Fine-tune on CV11 - 10.50 (CV11) (-49%)

• Zeroshot - 10.4 (Google Fluers)
• Fine-tune on CV11 - 9.26 (Google Fluers)

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0348	3.05	1000	0.1879	10.5033

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2