whisper-filter-CHIME6

This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.4559
• Wer: 188.5591

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use 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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.6668	0.1	500	1.6673	168.1719
1.9458	0.2	1000	1.6112	247.9148
1.2	0.3	1500	2.3070	235.5822
1.6182	0.4	2000	1.5156	168.1025
1.3974	0.5	2500	1.5074	180.2130
1.4654	0.6	3000	1.5212	131.2175
1.1705	0.7	3500	1.4534	143.9768
1.1804	0.8	4000	1.4870	174.3443
1.4669	0.9	4500	1.4604	193.9907
1.5055	1.001	5000	1.4559	188.5591

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0