noflm/whisper-base-ja-cv11

Whisper Base Japanese

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

• Loss: 0.6532
• Wer: 21.9918

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: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3273	3.02	1000	0.4225	20.8253
0.0923	7.0	2000	0.4643	21.2200
0.0164	10.02	3000	0.5403	22.9627
0.006	14.01	4000	0.5820	21.0861
0.0046	17.02	5000	0.5852	22.0728
0.0034	21.01	6000	0.6113	21.6623
0.0028	24.03	7000	0.6582	22.3266
0.0025	28.01	8000	0.6350	22.2332
0.0029	32.0	9000	0.6468	22.1098
0.0014	35.02	10000	0.6532	21.9918

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.8.1.dev0
• Tokenizers 0.13.2