whisper-tiny-german-V2-HanNeurAI

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5818
• Wer: 32.3327

This fine-tuning model is part of my school project. With limitation of my compute, I scale down the dataset from german common voice to shuffled 200k rows

Model description

Model Parameter (pipe.model.num_parameters()): 37760640 (37M)

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: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2054	0.08	1000	0.7062	39.0698
0.1861	0.16	2000	0.6687	36.4857
0.1677	0.24	3000	0.6393	35.6849
0.2019	0.32	4000	0.6193	34.4385
0.1808	0.4	5000	0.6103	33.8459
0.1697	0.48	6000	0.5956	32.8519
0.1468	0.56	7000	0.5884	32.7029
0.1906	0.64	8000	0.5818	32.3327

Framework versions

• Transformers 4.40.2
• Pytorch 2.3.0
• Datasets 2.19.1
• Tokenizers 0.19.1