A convolutive blind source separation algorithm based on tri-quadratic joint block diagonalization is proposed by utilizing the mutual-independence and quasistationarity properties of speech signals.The three subgroups of the undetermined parameters are iteratively estimated through minimizing the corresponding quadratic cost subfunctions alternatively.Analyses prove that the proposed algorithm has low computational complexity and robust convergence.Compared with the classical Jacobi-like methods

simulation results show that the proposed algorithm can improve both the global rejection level about 5dB and the Bark spectral distortion factor about 0.12 in average with faster convergence rate.