Authors:
Tomomi Abe
1
;
Mitsuharu Matsumoto
2
and
Shuji Hashimoto
2
Affiliations:
1
Major in Pure and Applied Physics, Waseda university, Japan
;
2
Waseda university, Japan
Keyword(s):
Noise reduction, speech enhancement, ε-filter, time-frequency domain.
Related
Ontology
Subjects/Areas/Topics:
Applications
;
Audio and Speech Processing
;
Digital Signal Processing
;
Multimedia
;
Multimedia Signal Processing
;
Pattern Recognition
;
Software Engineering
;
Telecommunications
Abstract:
A time-frequency ε-filter (TF ε-filter) is an advanced ε-filter applied to complex spectra along the time axis. It can reduce most kinds of noise while preserving a signal that varies frequently such as a speech signal. The filter design is simple and it can effectively reduce noise. It is applicable not only to small amplitude stationary noise but also to large amplitude nonstationary noise. However when we consider the noise that varies much frequently along the time axis, TF ε-filter cannot reduce noise without the signal distortion. When we consider the noise where the neighboring frequency bins have similar powers such as impulse noise, we can reduce the noise by using ε-filter applied to the complex spectra not along the time axis, but along the frequency axis. This paper introduces an advanced method for noise reduction that applies ε-filter to complex spectra not only along the time axis but also along the frequency axis labeled cross TF ε-filter. We conducted the experiments
utilizing the sounds with stationary, nonstationary and natural noise.
(More)