whisper-large-v3-turbo-sandi-train-1-pure-transcript-32

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the ntnu-smil/sandi2025-ds dataset. It achieves the following results on the evaluation set:

• Loss: 0.7791
• Wer: 18.5202
• Cer: 13.1470
• Decode Runtime: 188.5370
• Wer Runtime: 0.1495
• Cer Runtime: 0.2889

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: 7e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.98) and epsilon=1e-06 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 732

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Decode Runtime	Wer Runtime	Cer Runtime
0.9895	0.1667	122	0.8136	19.0375	13.4424	222.9064	0.1748	0.3402
1.1322	1.1667	244	0.7851	18.5866	13.1695	216.8919	0.1753	0.3360
0.5149	2.1667	366	0.7753	18.4884	13.1536	195.2818	0.1501	0.2897
0.3311	3.1667	488	0.7736	18.4361	13.0973	188.5320	0.1554	0.2902
0.8447	4.1667	610	0.7786	18.4750	13.1144	197.2527	0.1534	0.2967
0.9898	5.1667	732	0.7791	18.5202	13.1470	188.5370	0.1495	0.2889

Framework versions

• PEFT 0.15.2
• Transformers 4.52.2
• Pytorch 2.8.0.dev20250319+cu128
• Datasets 3.6.0
• Tokenizers 0.21.1