Abstract: The traditional Hilbert-Huang transform (HHT) based on the envelope line calculation of the cubic spline interpolation has several drawbacks such as overshooting and endpoint oscillation. To solve this problem, a new algorithm based on Gaussian process regression is proposed to compute the envelope. Optimizing the parameters of Gaussian process regression is discussed, with improved generalization ability and interpolation precision. The algorithm effectively resolves the false frequency and endpoint oscillation problems. Gait data processing indicates that the problem of spurious frequency and endpoint oscillation can be preferably improved with a Gaussian process, and distortion of mode function was decreased.