Speech hiding is a powerful information protection mechanism in real-time communication systems. By hiding one secret speech into the cover speech, we can get a stego speech, which sounds meaningful and indistinguishable from the original cover speech. Therefore, even if attackers intercept the audio packets on Internet, they would not notice that there is another speech hidden inside those packets. In this thesis, we propose a scheme for speech hiding in a real-time communication system such as voice over Internet Protocol (VoIP).