linshoufan/linshoufanfork-whisper-small-nan-tw

Whisper Small Taiwanese

This model is a fine-tuned version of openai/whisper-small on the Common Voice 16.1 and the Common Voice 15.0 datasets. It achieves the following results on the evaluation set:

• Loss: 0.3736
• Cer: 29.6973

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
0.7938	0.16	500	55.8341	0.7768
0.5845	0.32	1000	41.1522	0.5947
0.459	0.48	1500	37.6183	0.5132
0.3512	0.64	2000	35.4047	0.4709
0.3758	0.8	2500	33.5778	0.4363
0.3191	0.96	3000	32.6110	0.4216
0.2295	1.13	3500	0.4261	32.4977
0.1806	1.29	4000	0.4085	31.9909
0.16	1.45	4500	0.3913	31.1708
0.1603	1.61	5000	0.3836	30.3841
0.1343	1.77	5500	0.3784	30.1574
0.1265	1.93	6000	0.3736	29.6973

Framework versions

• Transformers 4.39.3
• Pytorch 2.1.2
• Datasets 2.18.0
• Tokenizers 0.15.2