Estimating the cross-correlation power spectra of cosmic microwave background (CMB), in particular, the T B and EB spectra, is important for testing parity symmetry in cosmology and diagnosing insidious instruments. The Quadratic Maximum Likelihood (QML) estimator provides the optimal estimates of power spectra, but it is computationally very expensive. The hybrid pseudo-C estimator is computationally fast but performs poorly on large scales. As a natural extension of previous work (Chen et al. 2021), in this article we present a new unbiased estimator based on the Smith-Zaldarriaga (SZ) approach of E-B separation and scalar QML approach to reconstruct the cross-correlation power spectrum, called QML-SZ estimator which allows us to use a scalar QML estimator to obtain the cross-correlaor. Our new estimator relies on the ability to construct scalar maps, within power spectrum. By reducing the pixel number and algorithm complexity, the computational cost is nearly one order of magnitude smaller and the running time is nearly two orders of magnitude faster in the test situations.