Hi there
Blind dereverberation is a very difficult task. I have spent the last
years of my research working on this topic.
Even though several approaches have been proposed, a possible
discrimination into two classes can be accomplished by considering
whether or not the inverse IR needs to be estimated. In fact, all
de-reverberation algorithms attempt to obtain dereverberation by
attenuating the IR effects or by undoing it. In a simplistic view, one
approach tries to alleviate the “symptoms” of the signal degradation,
while the other attempts to address its “cause”. Due to the spatial
diversity and temporal instability that characterize the IRs, the first
class of algorithms can offer, at the current state, more effective
results in practical conditions. However, the algorithms belonging to
the second class can potentially lead to ideal performances. It must be
noted that practical de-reverberation is still largely an unsolved problem.
“Reverberation suppression methods” are based on diverse set of
techniques such as: beamforming, spectral subtraction, temporal
envelope, LPC enhancement .
“Blind reverberation cancellation methods” can be distinguished into two
sub-classes: the techniques that are based on the IR blind estimation
followed by its inversion and the ones that attempt to directly estimate
the inverse system. While the first methods have the benefit of
providing the access to the IR estimation, and this is of interest for
the extraction of many acoustic parameters, the calculation of the
inverse system is not trivial even in the non blind case and it might
lead to inaccuracies. Therefore, it is probably more consistent for the
de-reverberation purpose to achieve a direct estimation of the inverse
system.
It is worthy to notice that blind reverberation cancellation and
suppression methods can be combined to offer hybrid strategies.
Even if cesptrum based techniques are popular in speech recognition,
they generally perform poorly since, speech and the acoustics cannot be
clearly separated in the cepstral domain. Furthermore, even if this
separation could be achieved, estimation of the clean speech signal is
still problematic since all phase information are removed. Therefore,
cepstral based methods can only be successfully applied in simple
reverberation scenarios (i.e. signal degraded by simple echoes).
you can find further information on:
www.dereverberation.org
Hope this is of help
Massimiliano Tonelli
Il 28/07/2011 11:24, Jerry ha scritto:
Alexandros, I haven't looked at your dissertation but it sounds very cool. Is
cepstral processing appropriate for removing reverberation? Just wondering.
Jerry
On Jul 21, 2011, at 10:08 PM, Alexandros Tsilfidis wrote:
Dear Kenneth,
Dereverberation is still an open and largely complicated research issue.
Depending on the application context (and on your needs) there are many methods
that you can chose from. However, from the mathematical point of view blind
dereverberation in realistic scenarios is a hard (or impossible) to solve blind
deconvolution problem and most methods make assumptions that produce
significant processing artifacts. You can download my PhD thesis on
dereverberation here:
http://www.wcl.ece.upatras.gr/audiogroup/alexandros/publications/Tsilfidis_PhD.pdf
where you can find a short literature review in p. 31-37.
Moreover, if interested you can download some of my personal publications and
listen to the corresponding audio demos at my personal website:
http://www.wcl.ece.upatras.gr/audiogroup/alexandros/
If you provide me with some more details on the specific application (e.g. how
important is the perceptual quality of the final results? do you have any
(other) prior knowledge than the recorded sound file? is the computational
complexity an issue for you?) probably I would be able to point you out the
specific method(s) that (may) suit your needs.
I hope that was helpful!
Alexandros Tsilfidis, MPhil, PhD
Post Doc Researcher
Department of Electrical and Computer Engineering
University of Patras
Greece
On 22 Jul 2011, at 06:50, Kenneth Ciszewski wrote:
Any suggestions about the best way to remove reverb from a sound recording
(voice/speech)?
--
dupswapdrop -- the music-dsp mailing list and website:
subscription info, FAQ, source code archive, list archive, book reviews, dsp
links
http://music.columbia.edu/cmc/music-dsp
http://music.columbia.edu/mailman/listinfo/music-dsp
--
dupswapdrop -- the music-dsp mailing list and website:
subscription info, FAQ, source code archive, list archive, book reviews, dsp
links
http://music.columbia.edu/cmc/music-dsp
http://music.columbia.edu/mailman/listinfo/music-dsp
--
dupswapdrop -- the music-dsp mailing list and website:
subscription info, FAQ, source code archive, list archive, book reviews, dsp
links
http://music.columbia.edu/cmc/music-dsp
http://music.columbia.edu/mailman/listinfo/music-dsp
--
dupswapdrop