02-Biometrics-Lecture-2-Part2-2008-09-29

02-Biometrics-Lecture-2-Part2-2008-09-29 - Master SC...

Info iconThis preview shows pages 1–18. Sign up to view the full content.

View Full Document Right Arrow Icon
1 Biometrics http://scgwww.epfl.ch/courses Master SC – Information and Communication Security Dr. Andrzej Drygajlo Speech Processing and Biometrics Group Signal Processing Institute Ecole Polytechnique Fédérale de Lausanne (EPFL) Center for Interdisciplinary Studies in Information Security (ISIS)
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
2 Speaker Recognition Generalities Feature extraction Speaker models and templates Speaker recognition errors Speaker recognition systems Advantages and disadvantages of voice as biometric
Background image of page 2
3 Two Phases of Speaker Recognition
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
4 Distance Diagram Acoustic vectors Reference model t T 1 1 L d (1, L ) d (1,1) d(t,L ) d(T,L ) d (1, R ( 1 )) d ( t , R ( t )) d(T,R ( T )) Accumulated Distance (score) d(t ,1) d(T ,1) Local distance
Background image of page 4
5 Example Enrollment Process Speech Signal Feature Extraction
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
6 Example Verification Process Speech Signal Feature Extraction
Background image of page 6
7 Templates in Speaker Recognition Deterministic and Statistical Methods Text-dependent recognition Dynamic Time Warping ( DTW ) Hidden Markov Model ( HMM ) Text-independent recognition Vector Quantization ( VQ ) Gaussian Mixture Model ( GMM )
Background image of page 7

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
8 Dynamic Time Warping (DTW) Reference Test Accumulated Distance
Background image of page 8
9 Dynamic Time Warping (DTW) Reference Test Accumulated Distance Template (Acoustic vectors)
Background image of page 9

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
10 Feature Extraction - MFCC-based front-end FFT FFT FFT based spectrum Speech DCT DCT Log Log Mel scale triangular filters Acoustic vector Δ Δ 2 13 MFCCs + E 13 Δ MFCCs 13 ΔΔ MFCCs Template Short word: 2 s Acoustic vector: every 10 ms Size: 2000/10 · 40 = 8 kB
Background image of page 10
11 Simple example of dynamic time warping (DTW) where each grid point ( i, j ) is associated with a local distance d ( i, j ) and an accumulated distance [ ] (, ) (, ) m in ( 1 , ) , (, 1 ) , ( 1 , 1 ) Di j di j Di j Di j j =+ Local constraints Reference acoustic vectors Test acoustic vectors (A) (B) (C) Dynamic Time Warping (DTW)
Background image of page 11

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
12 Dynamic Time Warping (DTW) Acoustic vectors Frames Reference acoustic vectors Local distance Accumulated distance
Background image of page 12
13 Dynamic Time Warping (DTW) Using the DTW algorithm we can solve for the minimum accumulated distance at each (allowed path) grid point giving: :[ ( ), ( )], 1,2,. .., wi k j k k K = (1) 1, 1 () , ij iK I jK J == 1 {[() ] , [() ] } () (, ) K k w dTik Rjk gk DT R Ng = = ( ) weighting function ( ) normalization factor
Background image of page 13

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
14 Local and accumulated distances Local distances Accumulated distances D %
Background image of page 14
15 Search Zones 1/ 2 pente globale 2/1 ≤≤ RA ij 1 () ( ) K k N Normalization gg k = = () [() ( 1 ) ] [() ( 1 ) ] gk ik ik jk = −− + Ng I J = + (, ) (, ) w DIJ DT R N g = %
Background image of page 15

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
16 Local Distance Three properties of local distance: A local distance measure used can have only one property valid (1). Local distance can be made symmetric: For vectors of K elements, the Hölder norm is as follows: 1) ( , ) 0 si ; 0 2) 3 ) (, ) (,) dxy x y dxx dxy dyx dxy dxu duy >≠ = = ≤+ sym 2 dx y + = 1/ 1 p K p pk k p k y x y xy = ⎡⎤ =− = ⎢⎥
Background image of page 16
17 Spectral Distance Logarithmic difference between two spectra:
Background image of page 17

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 18
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 06/25/2009 for the course MATH MAT 400 taught by Professor Jamespotvein during the Fall '08 term at University of Toronto.

Page1 / 53

02-Biometrics-Lecture-2-Part2-2008-09-29 - Master SC...

This preview shows document pages 1 - 18. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online