Make it possible to suggest words from sound similarity.

[cbrew]

A gensim model allows similarity queries from word embeddings. We want to be able to do this
directly using sound similarity.