Lecture "Pattern Recognition and Machine Learning"

Basic Information

Lecturers:   Gerhard Schmidt (lecture) and Tobias Hübschen (exercise)
Room:   F-SR-I
Language:   English
Target group:   Students in electrical engineering and computer engineering
Prerequisites:   Basics in system theory
Contents:  

In this lecture the basics of speech, audio, and music signal processing are treated. Often schemes that are based on statistical optimization are utilized for these applications. The involved cost function are matched to the human audio perception.

Topic overview:

  • Preprocessing to reduce signal distortions
    • Noise reduction
    • Beamforming
  • Speech and speaker recognition
    • Fundamentals of speech generation
    • Feature extraction
    • Gaussian mixture models (GMMs)
    • Artificial Neural Networks (ANNs)
    • Hidden Markov models (HMMs)
    • Recognition of speech and speakers
  • Enhancement of signal playback
    • Extending the bandwidth of speech signals

 

News

Starting this winter semester, the exercise will be extended by practical tasks such as training a neural network using Python and TensorFlow.  For this practical part of the excercise you are recommended to bring your own laptop. The exercise will be conducted as blocks, the schedule can be found below.

 

Schedule

The following schedule regarding lectures and excercises is preliminary and may be adapted during the semester. Each date refers to the time slot from 13:00 h until 16:30 h.

Date   Content
22.10.2019   Introduction
29.10.2019   Noise Suppression + Beamforming (part 1)
05.11.2019   Beamforming (part 2) + exercise
12.11.2019   Feature Extraction + Codebooks (part 1)
19.11.2019   Codebooks (part 2) + Bandwidth Extension
26.11.2019   Exercise
03.12.2019   Gaussian Mixture Models + Neural Networks (part 1)
10.12.2019   Neural Networks (part 2) + exercise
17.12.2019   Exercise
07.01.2020   Hidden Markov Models
14.01.2020   Speaker and Speech Recognition + exercise
21.01.2020   Exercise
28.01.2020   Student Talks

 

Lecture Slides

 

  Slides of the lecture "Introduction"
(Introduction, boundary conditions of the lecture, applications)
  Slides of the lecture "Noise Suppression"
(Noise suppression, dereverberation, speech reconstruction)
  Slides of the lecture "Beamforming"
(Fixed and adaptive beamforming, postfiltering)
  Slides of the lecture "Feature Extraction"
(Linear prediction, cepstrum, mel-filtered cepstral coefficients)
  Slides of the lecture "Codebook Training"
(K-means algorithm, LBG algorithms)
  Slides of the lecture "Bandwidth Extension"
(Model-bases approaches, evaluation)
  Slides of the lecture "Gaussian Mixture Models (GMMs)"
(Training with the EM algorithm, applications)
  Slides of the lecture "Hidden Markov Models (HMMs)"
(Efficient probability calculation, training of HMMs)
  Slides of the lecture "Speaker and Speech Recognition"
(Application of GMMs and HMMs, speech dialog systems)
  Slides of the lecture "Neural networks"
(Network types, training procecures)

 

Matlab Demos

  Matlab demo (GUI based) for adaptive noise suppression
  Matlab demo (GUI based) for linear prediction

 

Exercises

Please note that the questionnaires will be uploaded every week before the excercises, if you download them earlier, you won't get the most recent version.

  Questionnaire for the lecture "Noise Suppression"
  Questionnaire for the lecture "Beamforming"
  Questionnaire for the lecture "Feature extraction"
  Questionnaire for the lecture "Codebook training"
  Questionnaire for the lecture "Bandwidth extension"
  Questionnaire for the lecture "Gaussian Mixture Models"
  Questionnaire for the lecture "Hidden Markov Models"
  Questionnaire for the lecture "Speaker and speech recognition"
  Questionnaire for the lecture "Neural Networks"

 

Talks

At the end of the semester, each student will give a talk about a certain topic. The aim is both to give you the chance to work on a pattern recognition-related topic that interests you, and to improve your presentational skills. The talk is also a prerequisite for your admission to the exam. The talks should be held in English and should take ten minutes, plus 2.5 minutes of discussion and 2.5 minutes of feedback. Please write an email to This email address is being protected from spambots. You need JavaScript enabled to view it. to reserve your topic.

Below you can find the schedule of the talks.

Date   Room   Time   Topic   Presenter(s)
28.01.2020   F-SR I   xx:xx h   xxx   xxx

 

Exams

Below is the list of students with their exam dates. If you do not have a date for the exam yet please use the oral exam booking system on this website. You can find the booking system here.

Date   Time   Students (matriculation numbers)   Assessor
xx.xx.2020   xx:xx h   xxx   Tobias Hübschen

 

Website News

07.08.2019: Talk from Juan Rafael Orozco-Arroyave added.

11.07.2019: First free KiRAT version released - a game for Parkinson patients

25.06.2019: About 30 pupils from the Isarnwohld-Schule in Gettorf visited us.

02.05.2019: Christin Baasch finished sucessfully her defense on the evaluation of Parkinson speech.

30.11.2018: New student project on driver distraction added.

Recent Publications

   

J. Reermann, E. Elzenheimer and G. Schmidt: Real-time Biomagnetic Signal Processing for Uncooled Magnetometers in Cardiology, IEEE Sensors Journal, Volume 15, Number 10, Pages 4237-4249, June 2019, doi: 10.1109/JSEN.2019.2893236

Contact

Prof. Dr.-Ing. Gerhard Schmidt

E-Mail: gus@tf.uni-kiel.de

Christian-Albrechts-Universität zu Kiel
Faculty of Engineering
Institute for Electrical Engineering and Information Engineering
Digital Signal Processing and System Theory

Kaiserstr. 2
24143 Kiel, Germany

How to find us

Recent News

TensorFlow and Escape Rooms

The last two days (Monday and Tuesday, September 16 and 17, 2019), the whole DSS team did a course on TensorFlow and Keras. Since we already have interfaces of TensorFlow to our real-time tool KiRAT, it was time now to extend our knowledge on graphs and the corresponding training of weights, biases, etc. After two days of hard work we booked two escape rooms and solved the "secrets" hidden in ...


Read more ...