In this paper, we consider a semiparametric regression model involving both p-dimensional quantitative covariable X and categorical predictor Z, and including a dimension reduction of X via K indices X′βk. The dependent variable Y can be real or q-dimensional. We propose an approach based on SIRα and Pooled Marginal Slicing methods in order to estimate the space spanned by the βk's. We establish √n-consistency of the proposed estimator. Simulation studies show the numerical qualities of our estimator.