whisper-newts

This model is a fine-tuned version of trsree/whisper-newts on the Medical STT dataset 1.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0594
• Wer: 1.4166

Model description

Fine Tuned on openai whisper model for audio transcription with cardiology and other medical datasets.

Intended uses & limitations

ASR application, for automatic speech recognition and transcription.

Training and evaluation data

Cardio and general medical terminology datasets.

Training procedure

Supervised training with labelled datasets, Hugging Face transformers library, Jupyter python notebook

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• 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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0024	9.0090	1000	0.0574	1.3237
0.0001	18.0180	2000	0.0559	1.3934
0.0002	27.0270	3000	0.0584	1.4166
0.0	36.0360	4000	0.0594	1.4166

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0