marccgrau/whisper-small-init

Whisper Small German SBB

This model is a fine-tuned version of openai/whisper-small on the SBB Dataset 29.11.2022 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0151
• Wer: 0.8658

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8659	10.0	50	0.6119	6.4935
0.2183	20.0	100	0.0727	5.1948
0.0002	30.0	150	0.0168	0.8658
0.0001	40.0	200	0.0159	0.8658
0.0	50.0	250	0.0155	0.8658
0.0	60.0	300	0.0154	0.8658
0.0	70.0	350	0.0152	0.8658
0.0	80.0	400	0.0151	0.8658
0.0	90.0	450	0.0151	0.8658
0.0	100.0	500	0.0151	0.8658

Framework versions

• Transformers 4.25.1
• Pytorch 1.12.1
• Datasets 2.7.1
• Tokenizers 0.12.1