kimbochen/whisper-small-zh-tw

Whisper Small Traditional Chinese

This model is a fine-tuned version of kimbochen/whisper-small-zh-tw on the mozilla-foundation/common_voice_11_0 zh-TW dataset. It achieves the following results on the evaluation set:

• Loss: 0.4334
• Wer: 32.0420

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: 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: 1200
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0066	2.05	400	0.3743	32.9100
0.0084	5.03	800	0.3787	33.4171
0.0098	8.01	1200	0.3979	33.2481
0.0019	10.06	1600	0.4084	32.3116
0.0008	13.04	2000	0.4334	32.0420

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu116
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2