A transient vibration signal is generally induced by impact between parts in a mechanical system, so transient signal extraction is significant for the condition judgment and fault diagnosis of the machine. Considering that the human auditory system possesses the ability to segregate a novel burst signal stream in a multi-source sound field, a method for extracting the transient vibration signal component is proposed based on the mechanism of auditory stream segregation. First, band-pass filtering is used to obtain the time-frequency distribution of the analyzed signal. Then, the phase is adjusted using inverted filtering in every filter channel. Based on the maximum and minimal value of twice envelope of each filter signal, two cues (synchronism and mutation degree) are calculated to be used as index to screen out the time frequency segments that are relevant to the transient signal component. Finally, these time-frequency segments are added together to generate the waveform of the transient signal. Numerical simulation and the analysis of signals measured from the gear box show that the proposed method can extract a transient signal component from a powerful background signal with high accuracy of onset time, and has utilization potential.