search

Ant-Brain / EfficientWord-Net

OneShot Learning-based hotword detection.
https://ant-brain.github.io/EfficientWord-Net/
Apache License 2.0
215 stars 34 forks source link

Release the training code of new model. #34

Open yongliang-wu opened 1 year ago

yongliang-wu commented 1 year ago

Hello, the new model resnet50 works really good for me, but the size is still too big to deploy at edge device. Would you please release the training code and the new dataset link?

TheSeriousProgrammer commented 1 year ago

Thanks for the kind words. Here is the training code https://github.com/Ant-Brain/EfficientWord-Net-Trainer.git I have not found time to clean it yet , but here it is . The starting point of the code is simple_trainer.py

It is in our plans to create a MCU compatible version of the same. It would be helpful if you could share some of your insights in creating a trimmed down MCU compatible version of the same

yongliang-wu commented 1 year ago

Ok, I have found the code. But would you mind to provide the dataset? And maybe I can help u to clean them if I can train it.

TheSeriousProgrammer commented 1 year ago

The dataset is quite big around 100GB will try to upload it by today

TheSeriousProgrammer commented 1 year ago

a sad update. The dataset was present in my academic instance and it was taken down. will try to recreate it with the available scripts

TheSeriousProgrammer commented 1 year ago

You can make use of this for now , ml commons have uploaded the dataset with train test and validation in huggingface https://huggingface.co/datasets/MLCommons/ml_spoken_words/tree/main/data/wav/en

yongliang-wu commented 1 year ago

ok, thanks!