vumichien/whisper-large-v2-mix-jp

openai/whisper-large-v2

This model is a fine-tuned version of openai/whisper-large-v2 on the vumichien/preprocessed_jsut_jsss_css10_common_voice_11 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2284
• Wer: 7.6453
• Cer: 4.7187

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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
0.1912	0.55	1000	0.1828	11.2314	7.0357
0.1329	1.1	2000	0.1618	9.4172	5.9028
0.0912	1.65	3000	0.1616	8.9257	5.4711
0.0576	2.2	4000	0.1664	8.5861	5.3055
0.0449	2.74	5000	0.1642	8.4510	5.2930
0.02	3.29	6000	0.1799	8.1537	5.0354
0.019	3.84	7000	0.1801	8.125	5.0827
0.0067	4.39	8000	0.2003	7.8412	4.8133
0.006	4.94	9000	0.2071	7.5811	4.7023
0.0022	5.49	10000	0.2284	7.6453	4.7187

Framework versions

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