Exploring Quality and Generalizability for Parameterized Neural Audio Effects