jethrowang/whisper-tiny_tat_vanilla

Whisper Tiny Taiwanese (vanilla)

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: 0.9557
• Cer: 21.7372

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: 681
• training_steps: 6810
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.523	0.9985	681	0.7177	29.0462
0.3561	1.9971	1362	0.6283	24.2773
0.2406	2.9956	2043	0.6268	23.1643
0.1598	3.9941	2724	0.6796	22.8912
0.1	4.9927	3405	0.7482	23.3539
0.0618	5.9912	4086	0.8209	22.8447
0.039	6.9897	4767	0.8669	22.3618
0.0182	7.9883	5448	0.9197	22.4326
0.012	8.9868	6129	0.9375	21.9010
0.0085	9.9853	6810	0.9557	21.7372

Framework versions

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