marccgrau/whisper-small-allSNR

Whisper Small German SBB all SNR

This model is a fine-tuned version of openai/whisper-small on the SBB Dataset 05.01.2023 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0274
• Wer: 0.0209

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: 800
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3449	0.36	100	0.2161	0.0380
0.0653	0.71	200	0.0265	0.0178
0.0315	1.07	300	0.0291	0.0197
0.0194	1.42	400	0.0273	0.0197
0.0141	1.78	500	0.0278	0.0197
0.0088	2.14	600	0.0265	0.0159
0.0055	2.49	700	0.0273	0.0209
0.0047	2.85	800	0.0274	0.0209

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.1
• Datasets 2.8.0
• Tokenizers 0.12.1