A two-step approach to blind deconvolution of speech and sound sources in the time domain