vumichien/whisper-medium-mix-ja

openai/whisper-medium

This model is a fine-tuned version of openai/whisper-medium on the vumichien/preprocessed_jsut_jsss_css10_common_voice_11 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2836
• Wer: 8.7213
• Cer: 5.4698

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: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.1106	1.1	1000	0.1827	10.3480	6.4784
0.0487	2.2	2000	0.1799	9.4764	5.9127
0.0243	3.29	3000	0.1950	9.2111	5.8069
0.0106	4.39	4000	0.2113	8.9713	5.5756
0.0054	5.49	5000	0.2325	8.6470	5.4041
0.0031	6.59	6000	0.2462	8.7078	5.4409
0.0014	7.68	7000	0.2608	8.7145	5.4849
0.0009	8.78	8000	0.2695	8.6301	5.3876
0.0004	9.88	9000	0.2794	8.6064	5.3528
0.0003	10.98	10000	0.2836	8.7213	5.4698

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2