fast hermite transform?

I was talking to @dlfivefifty over at ApproxFun.jl where they have an implementation of the hermite transform using gausshermite quadrature. While this is clearly a reliable choice, is there any scope for implementing an O(N*logN) (i.e. faster than O(N^2)) method as described in
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2630232/#!po=88.5000
?
For band limited functions there is a factorization using recursion the relation, akin to FFT vs basic FT. I don't know it well enough to know how practical it is to implement. On the surface it appears an O(N*logN) method may be possible for any orthogonal polynomials with two-term recursion...

EDIT: it seems that the exact two-term recursion factorisation is numerically unstable, and in practice the transform is implemented as an approximate Hermite-Newton-Cotes transform on a linear grid. Probably my question should be closed here.

The issue names no repository files or tests. Start by reviewing the existing Gauss-Hermite implementation and the linked paper, then compare the proposed O(N*logN) approach with the numerical-stability concerns in the edit. Done would require a decided, practical implementation scope, but the author suggests closing the question.