yuweiiizz/whisper-small-taiwanese-hanzi

Whisper Small Taiwanese - hanzi

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

• Loss: 0.2705
• Cer: 16.2525

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3.2

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
0.5463	0.4	1000	28.5221	0.5585
0.4142	0.8	2000	24.6764	0.4017
0.2268	1.2	3000	19.0744	0.3413
0.2183	1.6	4000	18.1837	0.3093
0.2223	2.0	5000	0.2952	17.1932
0.125	2.4	6000	0.2840	17.1016
0.118	2.8	7000	0.2734	16.8893
0.0678	3.2	8000	0.2705	16.2525

Framework versions

• Transformers 4.40.2
• Pytorch 2.1.2
• Datasets 2.19.1
• Tokenizers 0.19.1