kimbochen/whisper-small-jp

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: 0.2543
• Wer: 13.7687

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: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2515	1.06	200	0.2881	16.9442
0.2212	2.12	400	0.2616	14.6884
0.0774	4.04	600	0.2543	13.7687
0.0564	5.09	800	0.2731	13.9769
0.0221	7.01	1000	0.2814	13.9700

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu116
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2