AR source modeling based on spatiotemporally diverse multichannel outputs and application to multimicrophone dereverberation