DIGITAL WATERMARKING OF AUDIO SIGNALS USING A PSYCHOACOUSTIC AUDITORY MODEL AND SPREAD SPECTRUM THEORY * By: Ricardo A. Garcia *Research done at: University.

DIGITAL WATERMARKING OF AUDIO SIGNALS USING A

PSYCHOACOUSTIC AUDITORY MODEL AND SPREAD SPECTRUM

THEORY*By:

Ricardo A. Garcia

*Research done at:University of Miami School of Music 1999

MIT Media LabMachine Listening Group

Objectives:

• Design an algorithm and implement a system capable of embedding digital watermarks into audio signals

• Use spread spectrum techniques to generate the watermark.

• Use a psychoacoustic auditory model to shape the watermark

Watermark characteristics:

• Not perceptible (transparent)

• Resistant to degradation– Removal attempts– Transmission by analog/digital channel– Sub-band coders

• Original audio is not required in recovery

Conference Overview:

1. a) Psychoacoustic Auditory Modelb) Noise shaping (watermark embedding)

c) Spread Spectrum watermark generation

2. Developed system

3. Examples and System Performance

4. Conclusions

a) PSYCHOACOUSTIC AUDITORY MODEL

• Simultaneous frequency masking

• Calculate an approximated masking threshold T(z) - frequency holes -

0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 2 2 2 4c ritic a l ban d ra te [B ark s]z

6 0

4 0

2 0

0Exc

itatio

n le

vel [

dB]

Psychoacoustic Auditory Model

F F T P o w er S p ec tru m

E n erg y p er C r itica l B a n d

S p re a d M a sk in gA c r o ss

C r it ic a l B a n d s

M a s k in gT h re sh o ldE s tim a te

N o ise S h a p in g

IF F T

s(t) S (j ) S p (j )

S p z(z )

B (z)

S m (z)

T (z)

N (j )

O U T (j )ou t(t)

Masking Threshold T(z)

b) NOISE SHAPING

• Replace components below masking threshold with components from a broadband noise-like signal (watermark)

• Level of the watermark below threshold

• Each critical band has its own scaling factor

Noise Shaping

c) SPREAD SPECTRUM

• Communication system– Uses all the available spectrum (broadband, noise-like)

– Each channel use an orthogonal code– All other channels appear as “noise”

F R E Q U E N C YT IM

E

CO

DE

F R E Q U E N C Y T IME

CO

DE

F R E Q U E N C YT IM

E

CO

DE

TDMAFDMA

CDMAspread spectrum

T ra n sm itte r R e c e iv e r

Ja m m e r

• Information = data sequence (watermark) • Jammer = music signal (after auditory model)

Direct Sequence SpreadingUncoded Direct Sequence Binary Phase Shift Keying

Uncoded DS/BPSK

P NG ene rator

B P S KM o d u la to r

d (t) x (t)

c (t)

s (t)

• Data sequence (watermark)

• Modulator (fo)

• PN sequence

Uncoded DS/BPSK

De-Spreading and Data Recovery

P NG en erato r

x(t)

c(t)

J(t)

y(t) r(t) ri d t

bT

0

Tran sm issio nC h ann el

)cos(2

0tTb

Coded DS/BPSK

• Transmitter:– Repeat Code (m) – Interleaving

• Receiver:– De-interleaving– Decoder (decision rule)

,,,,,,,

,,,,,,,

,,,,

2111332

1321321

dabcadb

cbbbaaa

dcba

2. PROPOSED SYSTEMTransmission: watermark generation and embedding

W AT E R M A R KG E N E R AT IO N

C o d ed D S /B P S K

P S Y C H O A C O U S T ICA U D IT O R Y

M O D E L

W AT E R M A R KS H A P IN G

A N D E M B E D D IN G

1 0 11 0 ...1 0 0 1

W A T E R M A R K(b it s tre a m )

A U D IO

W A T E R M A R K E DA U D IO

T (z ) T R A N S M I S S I O NC H A N N E L

PA R A M E T E R S

Reception: watermark recovery

T (z ) r (t )A U D IT O R Y

M O D E L

D E -S P R E A D I N GA N D

R E C O V E R Y

A D A P T IV EH IG H

R E S O L U T IO ND E T E C T IO N

R E S ID U A LG E N E R AT IO N

H E A D E RG E N E R AT IO N

1 0 11 0 ...1 0 0 1

T R A N S M IS S IO NC H A N N E L

R E C O V E R E DW A T E R M A R K

PA R A M E T E R S

3. EXAMPLES

Original Audio

After Auditory Model

Residual

One watermark

Shaped watermark

Watermarked Audio

SYSTEM PERFORMANCE

• Survival over different channels– MPEG, Mini Disc, Two consecutive D/A - A/D, Analog

Tape, FM Stereo Radio, FM Mono Radio, FM Mono Radio (weak signal), AM Radio

– (next slide)

• Listening test– ABX test, 40 trials

• (-2 db, 24 correct id.), (-4 db, 19), (-6 db, 19)

MPEG LAYER 3 Level: -2 dB

4. CONCLUSIONS

• The perceptual quality of the audio signal was retained

• The watermark signal survives to different removal attacks (redundancy)

• Few parameters are needed at the receiver to recover the watermark

FURTHER RESEARCH

• Performance with different types of music

• Changes in the playback speed of the signal

• Bit error detection and recovery

• Optimal spread spectrum parameters

• Multiple watermark embedding

• Crosstalk interference

Contact Information

• Ricardo A. Garcia– Email: [email protected]

– Website: http://www.media.mit.edu/~rago