Laser voice detection is a non-contact remote information acquisition technology, which has advantages of highly sensitivity, long work distance and anti-environment disturbance ability. However, under the influnece of backscattering characteristic, the performance of this technology was related to the angle of project laser beam, the kinds of object and surface roughness. At worst, the effective signal would be buried in oblivion, which made the system invalid. An eliminating light intensity disturbance phase generated carrier (PGC) demodulation algorithm based on the PGC-differential-cross-multiplying algorithm (PGC-DCM), which was used to solve the problem of light intensity disturbance caused by backscattering characteristics. Firstly, a new method mathematical model with LabView system design software was bulit. The frequency response and the performance of eliminating light intensity were analyzed. Through this algorithm, the influence of light intensity parameters B, G, H, J₁ and J₂ were eliminated; Secondly, for different light intensities, the performance of PGC-DCM algorithm and the new algorithm was compared. The experimental results showed that the new algorithm can demodulate the voice signal well, especially in the weak light environment. The PGC-DCM demodulation algorithm cannot demodulate the voice signal under the condition of weak light intensity (such as B=0.5 mW); Finally, a long-distance speech acquisition experimental system was established, and experiments on the influence of different light intensities were carried out. The experimental results show that under the conditions of 10, 20, 30, 40 mW detection optical power, the system demodulates the output voice signal with the same amplitude, which eliminates the influence of different light intensities on signal detection.