Stochastic vibration spectrum always contains sudden changes and discontinuance.On a wavelet denoising process, the soft-threshold method will make the estimation signal ambiguous at the discontinuity point, while the hard-threshold method will cause pseudo-Gibbs phenomena around the signal's discontinuity point. Through analysis on the statistic model of the stochastic vibration spectrum, a noise statistic model of numeric field vibration spectrum is established, and the nonlinear relationship between the filtering threshold-value and the wavelet transform scale is derived theoretically for providing a base for adaptive-threshold wavelet transform denoising. Finally,an universal adaptive denoising algorithm for vibration spectrum estimation based on wavelet transform is proposed. Simulation results show that the theoretical analysis is correct and the algorithm is good.