Whisper-Base-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: 1.2994
• Wer: 161.8521

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
1.3384	0.1	500	1.4503	127.1459
1.7048	0.2	1000	1.4166	215.8906
1.0308	0.3	1500	2.1531	221.1193
1.4247	0.4	2000	1.3488	135.8060
1.1564	0.5	2500	1.3525	178.7732
1.3189	0.6	3000	1.3451	131.0596
0.9805	0.7	3500	1.3045	131.0353
1.0246	0.8	4000	1.3280	150.3903
1.3236	0.9	4500	1.2997	157.3243
1.3807	1.001	5000	1.2994	161.8521

Framework versions

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