Actually, thinking about it, the resilient border interpolation at synthesis is not unnecessary now given by proposed WBHSM+LPC framework. The reason it became unnecessary before is that the proposed RVPM modeled each voice pulse as its full impulse response, assuming a time-invariant filter for each pulse (although I suppose the EpR differential envelope may model some of the time-variance of the filter, but this breaks as transforms are applied). The new WBHSM+LPC model is a literal combination of WBHSM and LPC where WBHSM models the source signal. Since that signal is not filtered, each voice pulse is actually independent, and pulses and periods are truly equivalent. Because of that, we actually go back to modeling each individual pitch period, reintroducing the need for the synthesis border interpolation. For this, we can do the same thing we do in the analysis stage where we shifted the window such that the window starts coincides roughly with the glottal opening instant and only the right side of the window is interpolated. This becomes especially beneficial for the WBHSM+LPC synthesis since the left side is where the voice pulse actually is, while on the right side the glottis should be closed or almost closed.