ntnu-smil/whisper-large-v3-turbo-sandi-train-1-ex-transcript-32-merged

whisper-large-v3-turbo-sandi-train-1-ex-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: 0.8091
• Wer: 23.6178
• Cer: 16.7291
• Decode Runtime: 196.7120
• Wer Runtime: 0.1710
• Cer Runtime: 0.3247

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
1.011	0.1667	122	0.8504	22.9206	16.3318	202.5562	0.1746	0.3201
1.1598	1.1667	244	0.8221	23.5312	16.7176	201.3143	0.1769	0.3246
0.5759	2.1667	366	0.8085	23.5030	16.6848	200.5117	0.1738	0.3214
0.361	3.1667	488	0.8055	24.0572	17.0616	199.6581	0.1785	0.3278
0.966	4.1667	610	0.8080	23.3946	16.5553	198.9523	0.1761	0.3228
0.9531	5.1667	732	0.8091	23.6178	16.7291	196.7120	0.1710	0.3247

Framework versions

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