sage-bergerson/whisper-large-ume-erj-v2

Whisper Large UME-ERJ V2

This model is a fine-tuned version of openai/whisper-large on the UME-ERJ dataset. It achieves the following results on the evaluation set:

• Loss: 0.0568
• Wer: 0.0496

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: 5e-06
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• 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
0.7362	0.1143	200	0.1780	0.1274
0.167	0.2286	400	0.1095	0.0852
0.1248	0.3429	600	0.0959	0.0776
0.0999	0.4571	800	0.0833	0.0669
0.0919	0.5714	1000	0.0821	0.0703
0.0839	0.6857	1200	0.0703	0.0623
0.0749	0.8	1400	0.0686	0.0611
0.0747	0.9143	1600	0.0689	0.0597
0.0624	1.0286	1800	0.0646	0.0586
0.0516	1.1429	2000	0.0638	0.0553
0.0497	1.2571	2200	0.0593	0.0521
0.0462	1.3714	2400	0.0634	0.0556
0.0454	1.4857	2600	0.0588	0.0516
0.0455	1.6	2800	0.0596	0.0540
0.0432	1.7143	3000	0.0622	0.0526
0.0401	1.8286	3200	0.0572	0.0524
0.0437	1.9429	3400	0.0569	0.0529
0.0344	2.0571	3600	0.0568	0.0496

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0
• Datasets 2.21.0
• Tokenizers 0.19.1