Psychoacoustic Approaches to Audio Steganography

Audio Steganography

Cody A. RayECES 434 Final

December 11, 2009

Introduction

•What is steganography?

•Security through obscurity

•Basis of watermarking

Requirements

•Hide a text-based message inside an audio signal with minimal distortion of the signal as perceived by the human ear.

Approaches

•Least-Significant Bit (LSB) Method

•Time Domain Amplitude Modulation

•Psychoacoustic Model from MPEG-1 Layer I

Least-Significant Bit

•Prefix message with identifier string.

•Embed 1 bit of the message as every LSB

•Does not require the original audio track to recover the message.

LSB System DesignCopy Copy

Header Header &&

Data Data SizeSize

Convert Convert Txt→BiTxt→Bi

nn

Embed Embed IdentifiIdentifier,Len, er,Len,

MsgMsg

Write Write New New Wav Wav FileFile

Extract Extract Length Length

of of MessagMessag

ee

Check Check for for

IdentifiIdentifierer

Extract Extract LSB LSB from from

WavDaWavDatata

ConvertConvertBin→TxBin→Tx

tt

Embed Message

Recover Message

Wav File Format

Start Byte Len Name

0 12 RIFF Header

12 24Wav Format Subchunk

36 8Data

Subchunk (ID and Size)

44 * Data

LSB Implementation

•Binary conversion using de2bi, bi2de (Communications Toolbox)

•Read bytes into array. Use bitset(A, BIT, V)

•Identifier = 101010102 = 17010

•Extended ASCII ¬ not sign

•ANSI ª Feminine Ordinal Indicator

Mono Wav LSB

Stereo Wav LSB

LSB Results

MonoMono StereoStereo

InputInput

OutputOutput

Amplitude Modulation

•Time domain amplitude modulation

•Slice time into blocks using message size

•Embed 1 bit of the message per time slice

•Requires the original audio track to recover the message.

AM System DesignCalculaCalcula

te te Block Block SizeSize

Convert Convert Txt→BiTxt→Bi

nn

Slice Slice Time Time

BlocksBlocks

ScaleScaleEach Each SliceSlice

Slice Slice Time Time

BlocksBlocks

CalculaCalculate te

Block Block SizeSize

ComparCompare Mag. e Mag.

ofofSlicesSlices

Convert Convert Bin→TxBin→Tx

tt

Embed Message

Recover Message

AM System Design•Binary conv. using Ascii: dec2bin,

bin2dec

•Block size =⎣len(sig)/len(msg)⎦

•Embed: scale slice by 0.99 if ‘1’ or 0.98 if ‘0’

•Recover: check slice is greater than original scaled to 0.98 then ‘1’, else ‘0’

Mono Wav AM

Stereo Wav AM

AM Results

MonoMono StereoStereo

InputInput

OutputOutput

MPEG-Based Steganography

•Selective Fourier Domain Amplitude Modulation using Psychoacoustic Models

•Psychoacoustic Models based upon the ISO-MPEG Model 1 Layer I, fs=44.1 kHz

MPEG System Design

1.Calculate the power spectrum.

2.Identify the tonal and non-tonal components.

3.Decimate the maskers to eliminate all irrelevant maskers.

4.Compute the individual masking thresholds.

5.Compute the global masking threshold.

6.Determine the minimum masking threshold in each subband.

7.Shape the power of the message below the masking threshold.

Demonstration

>> hide(‘my secret’, ‘input.wav’, ‘output.wav’)

>> recover(‘output.wav’)

message=‘my secret’

Future Work

•Complete MPEG-based system

•Explore statistical methods for recovery without possession of original file

•Explore non-textual steganography (compute symbols with a radix base)