In this paper an innovative model reduction technique is applied to the Variational Theory of Complex Rays in order to take account of uncertainties over a mid/high frequency band. Several methods have been developed to solve mid frequency problems, one of them is the VTCR, in order to solve noise and vibration problems in the medium-frequency range. The principal features of VTCR approach are the use of a weak formulation of the vibration problem, which allows to consider automatically boundary condition between sub-domains, without using any auxiliary equation such as Lagrange coefficients, and the use of Herglotz wave functions to represents the vibrational field; those functions are an integral repartition of plane waves in all the direction, the unknowns of the problem are their amplitudes. In order to improve its performances over a frequencies range, Proper Generalized Decomposition has been successfully applied in past works. The PGD is a model reduction technique that leads to construction of separated variables representations of the solution of models defined in tensor product spaces. Thanks to PGD, it is possible to write an approximation of wave amplitude on a reduced base, separating the polar contribution θ from the frequency dependent contribution ω leading to inexpensive broad band calculation. The main idea of this paper is to extend this model reduction to take into account parametric uncertainty. Being VTCR a pure deterministic method the easiest way to tackle stochastic problems would be to apply non-intrusive techniques, like Monte-Carlo simulation. The advantage of this technique is that it only requires to do several times the deterministic problem. Despite its simplicity, this method leads to important computational costs if considering complex cases and in particular broad bands. For that reason in this paper, the Proper Generalized Decomposition is proposed to consider uncertain parameters over a model order reduction, in addiction to the previous space/frequency decomposition on the reduced space. An iterative algorithm, based on residue minimization, is proposed in order to find the triplets (space, frequency and stochastic parameter) witch compose the solution over a reduced space. Finally an example is provided to show method efficacy on a stochastic mid-frequency broad bands and to show its potentialities to tackle high frequency.