In this paper, we propose a novel method for identification of gear crack from the low-frequency modulated vibration signal. Hilbert transform is used to present the envelope of the modulated vibration signal to show the modulating frequency. The new adaptive joint time-frequency signal representation is proposed to extract the demodulated characteristic frequencies from the Hilbert transform envelope of the vibration signal. The adaptive spectrogram matches the meshing frequency and its harmonics, the coupling frequency, the carrier frequency, and their sidebands in the vibration signal by the optimized wavelet. The simulated and experimental vibration signals are used to test the proposed method. The results show the applicability and effectiveness of the proposed method for gear crack detection.