Can Restricted Boltzmann Machine(RBM) be used for identifying important features for supervised machine learning?

Question

I have a dataset containing acoustic features extracted from audio using pyAudioAnalysis

It has a total of 136 features. 1-34 are the values of a window.