Abstract: The IPSO-QNN is applied to decreasing noise of speech signal in the time domain,and the emphasis is put on the improvement of the learning algorithm of QNN(quantum neural network).Aiming at the inherent shortcomings of premature with particle swarm optimization(PSO),an improved PSO(IPSO) is presented.The new arithmetic has better optimization performance by adding random data to premature particles' speed and position to make the premature particles leave the local optimum.A more efficient speech signal filter based on IPSO-QNN is established by using IPSO in learning and training of the parameters of QNN,a experimental platform is established using Matlab software,and experimental results show that the new arithmetic make best use of faster quantum neural computation and the global optimization ability of PSO,so that the speech signal filter has good performance in decreasing noise.