Many natural phenomena, such as volcanic eruptions, involve complex multi-phase flows. These flows often feature a mix of compressible and incompressible behaviors, making their modeling particularly challenging. A widely adopted framework is provided by the Baer–Nunziato equations for compressible two-phase flow. We present a semi-implicit solver for this system, featuring a novel linearly implicit discretization for both the pressure fluxes and the relaxation source terms, while the nonlinear convective terms are treated explicitly. This formulation leads to a CFL-type stability condition on the maximum admissible time step only based on the mean flow velocity, rather than on the sound speed of each phase, so that the novel scheme works uniformly for all Mach numbers. Implicit terms are discretized with central finite differences on Cartesian grids, avoiding artificial numerical diffusion in the low-Mach regime, whereas shock-capturing finite volume schemes are employed for the convective fluxes to guarantee robustness at high Mach numbers. The discretization of nonconservative terms preserves moving equilibrium solutions, making the method well-balanced, while the asymptotic-preserving property ensure consistency in the low-Mach limit of the mixture model. Second-order accuracy in space and time is achieved through the IMEX time-stepping scheme combined with TVD reconstruction. The proposed method is validated through a series of benchmark problems spanning a wide range of Mach numbers, demonstrating both its accuracy and robustness.