whisper-base-zh-20230724 - 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: 0.4868
• Cer: 21.2071

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
0.348	1.55	1000	0.6190	30.1519
0.1375	3.1	2000	0.4988	23.8969
0.0741	4.65	3000	0.4735	22.7089
0.0348	6.2	4000	0.4643	21.9984
0.0211	7.75	5000	0.4688	22.1851
0.0102	9.3	6000	0.4738	21.3982
0.0076	10.85	7000	0.4762	21.3477
0.0049	12.4	8000	0.4820	21.3352
0.0044	13.95	9000	0.4859	21.1040
0.0036	15.5	10000	0.4868	21.2071

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3