Aimed at the problem that singular value decomposition for singlechannel is poor in the effect on strong noise, a noise cancellation method based on dualchannel singular value decomposition is proposed. Firstly, the method of singular entropy determination is used to preprocess the noise component of high order, then the position of the low order singular vectors of the noise is determined from the correlation of the dualchannel singular vectors, and finally, the residual singular values and corresponding singular vectors are reconstructed to obtain the optimized estimated noisefree signal. The simulation experiments show that by the proposed method compared with the conventional method under condition of the low SNR and the white noise environment, the SNR gain increases by 4.07 dB and waveform correlation coefficient increment of pure signal increases by 0.11. On the other hand, a voice signal contaminated by cockpit noise is chosen as the experimental object. Compared the proposed method with the adaptive noise cancellation method for dualchannel, the results show that the SNR gain increases by 4.83 dB, and the operation takes a 1.5 s shorten. In addition, the proposed method is not restricted by the noise type, and also has a good adaptability to the colored noise, the singlefrequency interference and even the mixed noise, and a broad application prospect.