• Moritz Wild
  • 15.10.2012
  • 15.04.2013
An important task in audio signal processing is the suppression of interferers and noise. Typical applications include hands-free communication or speech recognition for smartphones, PCs or future interactive TV systems. Since usually only one source is of interest, it is desirable that all other sources as well as background noise are suppressed in the acoustic front end.

For sufficiently stationary noise, a common procedure for noise suppression is the following: During pauses of the desired speaker, a noise PSD estimate can be obtained. This estimate can then be used for noise suppression, e.g., by means of a Wiener filter.

For transient noise (keyboard typing, clock ticking, …), however, the PSD changes rapidly. Therefore, different approaches have to be applied.

In this thesis, a recently published algorithm tackling transient noise by means of so-called diffusion filters [Talmon 2011] should be analyzed. As a first step, an estimate of the transient noise should be obtained. Then, this reference should be used for suppressing the noise. The performance of the algorithm should be analyzed for different types of transient noise. Moreover, different versions of the algorithm should be compared.

Well-documented and well-structured software is important. The thesis can be written in German or English.