wav2vec2phone-large-xlsr-jp-jdrt5N-demo

This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-xlsr-53-japanese on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3714
• Wer: 0.4730
• Cer: 0.5054

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: 16
• eval_batch_size: 16
• seed: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.5238	1.0	567	1.3532	0.8709	0.6208
1.2812	2.0	1134	0.8674	0.6835	0.5633
1.1329	3.0	1701	0.7105	0.6164	0.5564
1.0267	4.0	2268	0.6111	0.5775	0.5401
1.0415	5.0	2835	0.5505	0.5499	0.5482
0.9767	6.0	3402	0.4986	0.5210	0.5204
1.0392	7.0	3969	0.4655	0.5082	0.5194
0.9235	8.0	4536	0.4457	0.4989	0.5136
0.9511	9.0	5103	0.4201	0.4917	0.5106
0.8998	10.0	5670	0.4031	0.4869	0.5081
0.8883	11.0	6237	0.3920	0.4814	0.5107
0.856	12.0	6804	0.3834	0.4790	0.5094
0.8814	13.0	7371	0.3772	0.4761	0.5081
0.8352	14.0	7938	0.3737	0.4735	0.5052
0.9001	15.0	8505	0.3714	0.4730	0.5054

Framework versions

• Transformers 4.31.0
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3