Steganalysis of adaptive multi-rate speech with unknown embedding rates using clustering and ensemble learning