jethrowang/whisper-tiny_tat_topline

Whisper Tiny Taiwanese (topline)

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

• Loss: 1.0448
• Cer: 20.4492

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: 0.0001
• train_batch_size: 64
• 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: 1362
• training_steps: 13620
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.6757	0.9985	681	0.6823	34.9394
0.5051	1.9971	1362	0.6048	25.3576
0.3472	2.9956	2043	0.5862	23.4704
0.2461	3.9941	2724	0.6068	22.2289
0.1633	4.9927	3405	0.6434	22.5768
0.1086	5.9912	4086	0.7158	21.8098
0.0723	6.9897	4767	0.7615	21.8697
0.0478	7.9883	5448	0.8106	21.8095
0.0333	8.9868	6129	0.8494	22.0310
0.0244	9.9853	6810	0.8936	21.9217
0.019	10.9839	7491	0.9116	21.9380
0.0124	11.9824	8172	0.9579	21.3454
0.0096	12.9809	8853	0.9779	21.6642
0.006	13.9795	9534	0.9902	21.4863
0.004	14.9780	10215	1.0026	21.0375
0.0021	15.9765	10896	1.0137	20.5766
0.0017	16.9751	11577	1.0258	20.7390
0.0015	17.9736	12258	1.0373	20.4630
0.0019	18.9721	12939	1.0412	20.5229
0.0013	19.9707	13620	1.0448	20.4492

Framework versions

• Transformers 4.42.3
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1