I am trying to implement a Deep Belief Network for speech recognition. And hence, need the first layer of RBM to have gaussian visible units.
I used @Andrej Karpathy's matrbm (https://code.google.com/p/matrbm/) and made some changes to convert the Bernouli-Bernouli RBMs to a Gaussian-Bernouli one.
The changed version is available at http://pastebin.com/KTPHnizd
Am I doing anything wrong?