jethrowang/android_loss_CH_0.5_emb-whisper-tiny

Whisper Tiny Taiwanese Simulated Android

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: 0.7416
• Cer: 11.5605

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
0.395	0.9985	681	0.4720	20.0872
0.278	1.9971	1362	0.4360	15.0426
0.1826	2.9956	2043	0.4391	14.4518
0.1179	3.9941	2724	0.4633	14.0327
0.0738	4.9927	3405	0.4930	12.9611
0.0491	5.9912	4086	0.5340	13.3159
0.0352	6.9897	4767	0.5716	13.2433
0.0238	7.9883	5448	0.6001	12.9938
0.0175	8.9868	6129	0.6153	12.7738
0.0123	9.9853	6810	0.6434	12.8122
0.0098	10.9839	7491	0.6496	12.6103
0.006	11.9824	8172	0.6643	12.5145
0.0037	12.9809	8853	0.6877	12.3994
0.0024	13.9795	9534	0.7057	12.2726
0.0017	14.9780	10215	0.7134	11.9908
0.0007	15.9765	10896	0.7194	11.8031
0.0004	16.9751	11577	0.7303	11.6993
0.0001	17.9736	12258	0.7350	11.6502
0.0003	18.9721	12939	0.7383	11.5326
0.0001	19.9707	13620	0.7416	11.5605

Framework versions

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