Digital Signal Processing

Prof. Dietrich KlakowRahil Mahdian

2

Language

• Teaching: English

• Questions: English (or German)

• Slides: English

• Tutorials: one English and one German group

• Exercise sheets: most likely English

• Your solutions: English or German

• Exam sheet: both languages

• Your answers: English or German

3

Course-Homepage

See:

• www. lsv. uni-saarland. De

Contains:

- Slides

- Exercises

- Literature

4

Register before Tuesday noon, please!

5

Use your university e-mail!

Register byTuesday12:00!

6

Lecture

• Lecture:

• Monday 10:15-11:45 (12)

• Location: -1.03 in building 13

• Contact:• Rahil Mahdian:

• tel. 58129

• Rahil.Mahdian@lsv.uni-saarland.de

• Mahdian.t.r@gmail.com

Termin OK?

7

Exercises

• Exercises:

• Time: by arrangement (doodle)

• Location: will be announced on mailing list

• Tutors:• Ali Kanso

• Merlin Köhler

8

Exercises/Practicals

• Details will follow

• Two groups

• Time and date: by doodle

• Programming language:

– Matlab (recommended!)

– or Coordinate with us before you proceed.

9

Mailing List

• We will set a mailing list• All students + ME, AK, and DK will be on it

• Purpose:• Raise questions to everybody

• Discuss questions

• Announcements (!!! e.g. exam dates !!!)

• … everything else …

• For changes contact: (I will tell you later)• Dietmar.Kuhn@lsv.uni-saarland.de

10

Exam

• Written exam• 120 Minutes• Date: doodle

• Note: CuK master students can only take it as a corecourse if it was not mandatory in their bachelorsprogram

11

Prerequisites

• Some math

• Programming

• E.g. matlab ,C++, java, maple, ...

12

Rules of the Game

• In case you don't understand something:

1. Ask!!!

2. Ask!!!

3. Ask!!!

1. Introduction

14

Outline of the Lecture

1. Introduction: (3 lectures)i. DSP basics, Sampling, Convolution, Windowing, etc.

ii. Filtering (e.g., FIR, IIR, etc)

2. Signal representations, analysis, and modeling: (3/4 lectures)i. Transforms: DFS, DTFT, DFT, FFT

ii. Multi-resolution Analysis: T-F (Gabor), FilterBank, Wavelet

iii. Sparse Representations: DCT, Wavelet

iv. Signal modeling: LP, AR, ARMA, etc

3. Microphone Arrays (e.g., multi-sensor processing) – 1 lecture

4. [Filtering and Smoothing] : (1-lecture)

5. Feature Extraction (Speech, Image) – (1 lecture)

6. PCA (KLT), LDA, ICA/IVA (MIMO), etc – (2 lectures)

7. Statistical (& Adaptive) Signal Processing: (2/3 lectures)i. Wiener filter

ii. Spectral Subtraction

iii. LMS, RLS, Kalman filter

15

DSP – The main goal

A generic term for some techniques: e.g.,a. Filteringb. Spectrum analysisc. Compressiond. Separatione. etc

applied to digitally sampled signals.

16

Signal- Analog and Digital

17

Image Signal

Taken from prof.sKatsaggelos, NWU

18

Image Signal- resolution

Taken from prof.sKatsaggelos, NWU

19

Signal- Quantization

Taken from prof.sKatsaggelos, NWU

20

Multi-Spectral Imaging

Taken from prof.sKatsaggelos, NWU

21

Multi-Spectral Imaging

Taken from prof.sKatsaggelos, NWU

22

Feature Extraction from Images

Main features:•Color•Texture•Edges

23

Speech Signal - representations

24

Feature Extraction from Speech

Standard featureextraction fromspeech:Mel-Frequency-CepstralCoefficients

25

Intersection of DSP and scientific areas

26

Beamforming Blind Source Separation

Multi-sensor signal processing

27

Microphone Arrays

+

d τ=dc

sinθ

Use time delay toenhance signalfrom a certaindirection.

28

Spatial Filtering - Beamforming

29

Test Data xi

Basic Principle of Pattern Recognition

Feature Extraction

Classifier Model

1 ….

Feature Extraction

Training Data

Training Algorithm

2 n

30

Musical Genre Classification

?

Classical

Country

Rock

31

or Speaker Recognition

Speaker verification:is this Mary?

Speaker identification:who is speaking?

32

or Classification

• A simple introduction

• Nearest Neighbor Classifier

33

KL-Transform andLinear Discriminant Analysis

Find the optimalsubspace forfeature vectors

34

Linear Predictive Coding

Basic algorithm forspeech coding

35

Linear Filters

With “salt-and-pepper”noise

Gaussian3x3-kernel

blurring of the image use other filters (e.g. median filter)

36

Spectral Subtraction and Wiener Filter

From: http://www.tu-harburg.de/ft2/AktuForschungen/Bildverarbeitung/Bildverarbeitung.htm

Suppress noise

37

Literature

Applied Pattern Recognitionvon Dietrich W. R. Paulus, Joachim HorneggerViewegISBN: 3528355581Ca. 40 Euro

•Speech and image analysis•Software oriented•Signal processing•…

38

Literature

Digitale Sprachsignalverarbeitungby Peter Vary, Ulrich Heute, Wolfgang HessTeubner VerlagISBN: 3519061651Ca. 45 Euro

•Spectral subtraction•Wiener Filter•Microphone arrays•…

39

Literature

Spoken Language Processingby Xuedong Huang, Alex Acero, Hsiao-Wuen Hon,Xuedong Huang, Hsiao-Wuen HonPrentice HallISBN: 0130226165

• very comprehensive

40

Questionnaire

• Take a sheet of paper

• Please give me you opinion of thosequestions:• All topic covered that you expect?

• Is there a topic we could skip?

• What do you expect in terms of teaching style

• What should not happen in the lecture

Take yourself 5 minutes time