cankaracelebi/whisper-medium-tr-v4

Whisper Medium Tr - Can K V2

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

• Loss: 0.2285
• Wer: 15.4185

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.203	0.3448	1000	0.2255	19.4192
0.1602	0.6895	2000	0.2142	18.0448
0.0814	1.0343	3000	0.2087	17.5338
0.0761	1.3791	4000	0.2060	17.1558
0.0734	1.7238	5000	0.1998	16.5052
0.0335	2.0686	6000	0.2073	16.7283
0.0344	2.4134	7000	0.2066	15.9091
0.0338	2.7581	8000	0.2023	15.3709
0.0099	3.1029	9000	0.2211	15.6331
0.0097	3.4477	10000	0.2254	15.6008
0.0096	3.7924	11000	0.2254	15.3334
0.0022	4.1372	12000	0.2285	15.4185

Framework versions

• Transformers 4.43.3
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1