Denoising and block thresholding

21.10.2020 14:00 - 15:30

Luis Alberto Escudero (Acoustics Research Institute, Austrian Academy of Sciences)

It is widely known that time-frequency analysis (TFA) finds one of its
most successful applications in audio denoising. The core of a TFA-based
denoising method often consists of examining coefficients in a discrete
time-frequency representation (TFR). In some noise subtraction
techniques, such inspection is carried out singling out each
coefficient. However, this method often introduces an artifact called
"musical noise" in the reconstructed signal. Such artifacts are mostly
due to ignoring the redundant structure of a TFR in the filtering
process, and one way to reduce them is by filtering blocks instead of
individual coefficients.

In this talk, we will briefly review tight frames and Gabor multipliers,
followed by a presentation of how audio denoising can be achieved by
time-frequency block thresholding.


*Yu, Guoshen*; *Mallat, Stephane*; *Bacry, Emmanuel*. Audio denoising by
time-frequency block thresholding. /IEEE Trans. Signal Process./*56
*(2008), no. 5, 1830--1839.


Meeting ID: 818 7445 7687

Passcode: 969502

J. L. Romero, M. Ehler