An adaptive speech source separation algorithm under overcomplete-cases using Laplacian mixture modeling for mixture matrix estimation by adaptive EM-type algorithm in wavelet packet domain.