If your input is only ever a single word, that's right.

Usually people work with texts, or sequences of words. The embedding layer maps the sequence of words to a sequence of embedding vectors. It could be implemented as a sequence of one-hot encodings multiplied by the same W though.