The invention discloses a voice identification algorithm of a modular robot. The voice identification algorithm comprises the following steps: S1, performing interval sampling on a voice feature function f(x) to obtain m function values (X0(E), X1(E), L, Xm(E)); S2, converting the feature function f(X)={(X0(E), X1(E), L, Xm(E)} in a time domain into points in an m-dimensional Hilbert space by taking X0, X1, L and Xm as coordinates in the Hilbert space, and converting a voice sampling function family F(x) into a series of points in the Hilbert space; S3, analyzing a similarity relation among the series points of F(x) by taking the Hilbert space as a new feature space of a voice signal, and obtaining a mode identification module by applying a high-dimensional hypersphere covering method; S4, performing mode identification on the voice signal in the Hilbert space to complete voice identification. The voice identification algorithm of the modular robot has the most important characteristics that no FFT is provided, the complexity is low, and the algorithm is simple; the voice identification algorithm is suitable for a voice control system of a smaller-sized modular robot.