Green's functions recovering from ambient noise require strong hypothesis, including isotropic distribution of the noise sources. This ideal distribution is not achieved in most of the experimental applications, yielding to bias in the Green's function reconstruction. To minimize this bias, a spatio-temporal inverse filter is proposed. In this study, an experimental active seismic dataset is used to compute a directive noise field. The method is then implemented on this controlled dataset. The results indicate that the passive inverse filter can efficiently compensate for the noise wavefield directivity by manipulating the spatio-temporal degrees of freedom of a complex wave field.