whisper-base-zh-20230711 - au2a

This model is a fine-tuned version of openai/whisper-base on the some hakka audio dataset. It achieves the following results on the evaluation set:

• Loss: 0.4551
• Cer: 16.9978

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: 5e-06
• train_batch_size: 32
• 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: 500
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.4673	0.65	1000	0.6526	25.0548
0.2203	1.29	2000	0.4985	19.8459
0.1446	1.94	3000	0.4557	18.0026
0.0956	2.59	4000	0.4438	16.9676
0.0527	3.24	5000	0.4450	17.0998
0.0423	3.88	6000	0.4441	17.7797
0.027	4.53	7000	0.4474	16.9260
0.0177	5.18	8000	0.4515	16.5861
0.0165	5.83	9000	0.4537	16.8392
0.0129	6.47	10000	0.4551	16.9978

Framework versions

• Transformers 4.30.2
• Pytorch 1.11.0+cu113
• Datasets 2.13.1
• Tokenizers 0.13.3