jethrowang/whisper-tiny_tat-esc_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: 1.3924
• Cer: 32.8471

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: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• 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
0.3116	0.9985	681	0.9744	57.5703
0.1801	1.9971	1362	0.9761	37.9992
0.1094	2.9956	2043	1.0098	36.0103
0.0642	3.9941	2724	1.0710	34.1475
0.0353	4.9927	3405	1.1779	34.8229
0.0194	5.9912	4086	1.2733	34.6312
0.0086	6.9897	4767	1.3132	34.7455
0.0027	7.9883	5448	1.3640	33.1173
0.0009	8.9868	6129	1.3809	32.4291
0.0005	9.9853	6810	1.3924	32.8471

Framework versions

• Transformers 4.49.0
• Pytorch 2.0.0.post304
• Datasets 3.3.2
• Tokenizers 0.21.0