vumichien/whisper-small-ja

Whisper Small Japanese

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

• Loss: 0.4232
• Wer: 13.4679
• Cer: 8.6022

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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0167	7.0	1000	0.3066	13.6740	8.5733
0.0021	14.01	2000	0.3579	13.8733	8.7816
0.0006	21.01	3000	0.4025	13.5794	8.6173
0.0004	28.01	4000	0.4232	13.4679	8.6022
0.0004	35.01	5000	0.4319	13.4747	8.6213

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.12.1+cu113
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2