Computer-generated hologram (CGH) is recently expanding its application fields. However, the calculation cost is very high, in particular, in the generation of CGH streams for three-dimensional movies. This paper proposes a small-calculation-cost method to generate CGH streams based on a coherent neural network (CNN) that deals with complex-amplitude information with generalization ability in the carrier-frequency domain. After carrier-frequency-dependent learning, we can generate a CGH stream, by sweeping a virtual carrier frequency in the CNN, with neural interpolation thanks to the frequency-domain generalization. Experiments demonstrate a successful stream generation with 1/6 the conventional calculation time.