Abstract: On the basic of the undeterminate angle of discrete fractional fourier transform and the close relation between Bessel-Fourier moments and the corresponding scrambled signal, a security algorithm for digital speech forensics robust against desynchronization attacks was proposed. The scheme solved the problem of insecurity and incapable of tamper location for watermark schemes based on synchronization code. Watermark bits were generated by using frame number and discrete fractional fourier transform (DFRFT) coefficients, and embedded into amplitudes of Bessel-Fourier moments(BFMs) of scrambled signal. Theoretical analysis and experimental evaluation results showed that, comparing with the existing watermark schemes, the proposed scheme not only was inaudible and tolerance against common signal processing operations, but also had the excellent ability of tamper detection and improved the security of watermark system.