Accurate and Scalable Version Identification Using Musically-Motivated Embeddings