CNN-VAD
CNN-VAD copied to clipboard
Script and trained model for prediction on Desktop
Thanks for the great repository!
I am interested in a result movie that is robust to noisy enviroment. I want to apply and eveluate your method to various noise audio.
Scripts and models that work on the desktop, not the mobile, are desired. Is it possible to provide it?
Hi Kazuki
Thank you for looking at the repository. The model trained can be used on any device. I've attached the relevant resources which might be helpful for your application. The feature extraction module is written in MATLAB, but the other code is written in C. Our model was trained with a low-pass filter in the pre-processing stage so please be sure to use that filter as well, otherwise the results will be incorrect.
Trained Model: https://github.com/SIP-Lab/CNN-VAD/blob/master/CNN-VAD-iOS/CNN_VAD/frozen_without_dropout.pb
Loading saved model: https://www.tensorflow.org/tutorials/keras/save_and_load
Feature Extraction Code: https://github.com/SIP-Lab/CNN-VAD/blob/master/Training%20Code/Feature_Extraction.m