fs-w-he-base-en

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

• Loss: 1.0616
• Wer: 134.8528
• Cer: 128.2635

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: 16
• eval_batch_size: 8
• 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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
5.5657	4.5872	500	5.7291	440.3609	462.4528
1.368	9.1743	1000	1.8578	127.8727	122.6554
0.5271	13.7615	1500	1.0031	101.1871	87.2552
0.3551	18.3486	2000	0.7689	107.6448	90.9653
0.2646	22.9358	2500	0.7395	197.5309	237.3841
0.1357	27.5229	3000	0.7802	124.2165	120.3023
0.0493	32.1101	3500	0.8682	150.9497	148.3167
0.0112	36.6972	4000	0.9626	140.2659	152.1384
0.001	41.2844	4500	1.0325	135.9924	125.2662
0.0004	45.8716	5000	1.0616	134.8528	128.2635

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0