Can anybody explain how Embedding layers are used to represent the data ?

[Atralb]

Hi guys,

I am trying to understand the implementation and architecture of the network model.
There is an option to make use of Embedding layers to represent the midi files. I know their use as vector representation of words in a given vocabulary but I don't understand how the midi samples can take the role of words, since we don't want the generator to use the exact same vocabulary as in a language neural network right ? Otherwise, he won't create new note patterns thanks to learned low-level features.
I'm sure there is smth I don't quite get but I can't find what, so that's why I'm asking for your help :).

Thanks.