whisper-large-v3-turbo-common_voice_16_1-zh-TW-pissa

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the JacobLinCool/mozilla-foundation-common_voice_16_1-zh-TW-preprocessed dataset. It achieves the following results on the evaluation set:

• Loss: 0.5133
• Wer: 63.6656
• Cer: 23.5752

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: 0.0005
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0	0	2.7520	77.6125	20.7783
7.6982	0.9987	377	0.8744	87.9421	41.2804
5.1677	2.0	755	0.7499	82.5965	36.6407
3.3647	2.9987	1132	0.6433	76.8087	31.6068
3.4711	4.0	1510	0.6397	76.2460	30.2862
1.5694	4.9987	1887	0.5779	71.5434	27.5471
0.7951	6.0	2265	0.5664	71.3223	27.0600
0.4709	6.9987	2642	0.5492	68.8706	26.0131
0.116	8.0	3020	0.5427	66.7605	24.8104
0.0512	8.9987	3397	0.5298	66.1375	24.8632
0.0273	9.9868	3770	0.5133	63.6656	23.5752

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.4.0
• Datasets 3.0.2
• Tokenizers 0.20.1