Sidebar Menu

Seminar "Selected Topics in Speech and Audio Signal Processing"

 

Basic Information

 
 
Seminar overview  
  Lecturers:   Gerhard Schmidt and group  
  Semester:   Winter term  
  Language:   English or German  
  Target group:   Master students in electrical engineering and computer engineering  
  Prerequisites:   Fundamentals in digital signal processing  
  Registration
procedure:
 

If you want to sign up for this seminar, you need to register with the following information in the registration form

  • surname, first name,
  • e-mail address,
  • matriculation number,

Please note that the registration period starts 01.10.2020 at 8:00 h and ends 28.10.2020 at 23:59 h. All applications before and after this registration period will not be taken into account.

Registration will be possible within the before mentioned time under the following subsite - Seminar Registration.

During the registration process you will also choose your seminar topic. Only one student per topic is permitted (first come - first serve).

The registration is binding. A deregistration is only possible by sending an e-mail with your name and matriculation number to This email address is being protected from spambots. You need JavaScript enabled to view it. until Sunday, 18.10.2020 at 23:59 h. All later cancellations of registration will be considered as having failed the seminar.

 
  Time:   Preliminary meeting, Zoom, 04.11.2020 at 15:00 h
Written report due on 26.02.2021
Final presentations, Zoom, xx.xx.2021 at xx:xx h
 
  Contents:  

Students write a scientific report on a topic closely related to the current research of the DSS group. Potential topics, therefore, deal with digital signal processing related to speech and audio processing.

Students will also present their findings in front of the other participants and the DSS group.

 

 

Topics for WS 20/21

 
 
Topic title   Description  
  Speaker Separation in Reverberant Environments  

Speech enhancement and speech separation are two related tasks, whose main purpose is to extract either one or more target speech signals, respectively, from a mixture of other interfering sources. Speaker separation is a special case of speech separation, in which the mixture signal comprises two or more speakers. The aim of this seminar is to review the most recent signal processing and machine learning techniques applied for speaker separation especially those techniques which address the speaker separation problem in realistic reverberant environments.

 
  Statistical Measures for the Analysis of Evaluation and Prediction Scores  

In speech and audio processing, a common aim is to improve the quality of a disturbed signal. Consequently, it is desirable to check whether this aim was achieved by a new processing chain or algorithm. For this task studies with human subjects can be conducted, resulting in evaluation scores from which a statistically firm conclusion is to be derived. This seminar topic should gather information on statistical measures applicable to such statistical analyses. In addition, the analysis of prediction scores, which are produced by an analysis algorithm instead of human subjects, should be considered.

 
  Acoustic Artifacts
Imposed by
Speech Enhancement Systems
 

Communication systems such as ICC systems or communication units in firefighter breathing mask are implemented to enhance communication between two or more persons or parties. Often, such systems have to deal with several challenges at one time. Therefore, some unwanted effects may occur. The aim of this seminar is to review the mentioned systems regarding possible unwanted effects.

 
  Neural Network-based
Optimal Step-Size
Estimation
 

Normalized least mean squares (NLMS) based adaptive filters are used in many applications such as for echo cancellation in hands-free telephony. Here, a fast convergence and a good accuracy is mandatory for the system's overall performance. The speed of convergence depends on the so-called step-size, where a large step-size means a fast convergence and a small step-size leads to a slow convergence but a better steady state performance. Whenever a fixed step-size is chosen, a tradeoff has to be made. However, there are a lot of adaptive step-size approaches, which are mostly based on an estimation of the undisturbed error signal. Neural networks are versatile, so they are used for many different applications. In this work, neural network-based approaches for the optimal step-size estimation should be investigated by means of a literature research.