Detecting Steganography of Adaptive Multirate Speech with Unknown Embedding Rate