I'm not sure if that's the best way to describe them. IMO, NumPy/SciPy are probably best described as Python wrappers for a wide range of C and Fortran codes, but they add quite a lot of value on top of that. If you've ever used e.g. LAPACK or VODE directly, you'll probably have horrible memories that you prefer not to recall. NumPy and SciPy have a large contribution in that they unify (admittedly, neither is really a paragon of consistency, but...) a huge range of disparate libraries under a much more accessible interface.
To say NumPy or SciPy is just the sum of the libraries it uses is a bit like calling MATLAB a fancy interface for LAPACK. That really was what MATLAB was intended to be originally, but to reduce it to that is missing the appeal. As somebody who absolutely loathes MATLAB, even I wouldn't think that is fair.
The major feature of Python for me is that all its various open source modules for anything n-dimensional (matrices, image processing, GIS, machine learning, etc etc) all use the same data structure: the Numpy array. It's basically part of the standard library that is developed outside of it for pragmatic reasons. And with the speed of Fortran array operations.
That's something that only Matlab really has as well, but Python is much better in other aspects.
Which is why numpy and scipy are so full-featured and pretty damn fast.