LSTM on Ossian from CSTR-Edinburgh

Some modifications were brought to the initial project:

1. Getting two characters at a time during parsing :

"scripts/processors/Phonetisers.py"

2. Updated Theano flag and removing gpu bloking :

"scripts/util/submit.sh"

3. LSTM config file :

"scripts/merlin_interface/feed_forward_dnn_ossian_acoustic_model.conf"

4. LSTM were not implemented in original Ossian:

a) Storage in Ossian format is now possible:

"scripts/util/store_merlin_model.py"

b) Adding LSTM forward pass at test time

"scripts/processors/NN.py"

5. Removing VSM features that did not show to be useful

"recipes/naive_01_nn.cfg"