BrilliantCoolHuge/whisper-tiny-hi

Whisper Small Chinese - Bingcheng Hu

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: 1.7064
• Wer: 226.4151

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5
• training_steps: 400
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8417	1.08	20	1.4964	598.1132
1.2552	2.16	40	1.4901	367.9245
0.8195	3.24	60	1.4953	391.5094
0.6174	4.32	80	1.5091	475.4717
0.4594	5.41	100	1.5318	520.7547
0.489	6.49	120	1.5558	863.2075
0.4673	7.57	140	1.5719	663.2075
0.3976	8.65	160	1.5962	682.0755
0.3518	9.73	180	1.6160	623.5849
0.3043	10.81	200	1.6219	620.7547
0.2524	11.89	220	1.6505	598.1132
0.259	12.97	240	1.6543	329.2453
0.1696	14.05	260	1.6678	333.0189
0.1188	15.14	280	1.6746	329.2453
0.1366	16.22	300	1.6852	428.3019
0.1165	17.3	320	1.6979	262.2642
0.1062	18.38	340	1.7021	338.6792
0.0882	19.46	360	1.7047	313.2075
0.0891	20.54	380	1.7054	302.8302
0.0676	21.62	400	1.7064	226.4151

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 1.11.0+cu102
• Datasets 2.10.1
• Tokenizers 0.13.2