Aiming at the problem that the performance of time delay estimation algorithm based on sparse linear prediction is degraded under the environment with noise and reverberation

an improved linear prediction model is proposed. In order to obtain the effective pre-filtering of the microphone signal

the sparseness of the speech amplitude spectrum and the sparsity of the linear prediction vector are simultaneously introduced into the least squares criterion to construct the convex constrained linear prediction model. Then the split-Bregman iterative method is used to solve the model. Finally

the signal of prediction error is used to establish a time delay estimator which is based on the improved L2/L1-norm sparse linear prediction. The experimental results show that compared with the algorithms of GCC-PHAT and L2/L1-norm sparse linear prediction

the proposed algorithm has better performance of time delay estimation.