Recently, a few researchers have started to realize that in order for humanoid robots to move more naturally, it is necessary for them to incorporate a flexible spine in their robots. So far, nobody has come out with a solution which allows their spine robots to maintain balance in real-time. This paper presents a biologically inspired, hybrid CPG-ZMP controller for a single-legged, flexible spine belly dancing robot. Using only two control parameters, our robot can generate rhythmic and wave-like spine motions through the CPG component. By monitoring the torque at the robot's ankle, the ZMP component allows the robot to maintain balance in real-time. Unlike traditional ZMP-based controllers, no modeling of the robot's dynamics or explicit computations of the zero moment point are required. The resultant robot's motions emerge automatically in real-time through dynamic interactions between the robot, its neural network and the environment. Experimental results indicate that our controller has the potential to be applied to a new generation of flexible spine, biped walking humanoid robots.