This paper proposes a three layered model for audio steganography based on least significant bit replacement. The secret message to be transmitted is passed through two layers before it is embedded within the cover message in the third layer. The stego message is retrieved from the network by the receiver and the secret message is recovered by performing reverse operations in reverse order. The main focus of the paper is to make sure the confidentiality of the secret message. Implementation issues of the three layered model with respect to different parameters including capacity, transparency and robustness are discussed.