Source Separation

On Feb 27, 5:02 pm, "dbormpou" <dborm...@bio.auth.gr> wrote:
> Hello
>
> I am a biologist and i am trying to eliminate
> certain sounds from a stereo recording eg the sound of
> cars from the recording of a park, without loosing a
> lot in audio quality.
>
> I have some experience in audio and computers
> though i am not familiar with programming
>
> I need suggestions on some programms i can use
> for this purpose

Look up blind source separation or blind signal
separation.  It's research topic, usually confined
to particularly well defined sound sources, not
something for which you' are likely find a canned
application for your particular recording.

Hi Dimitris,

On 28 f&#4294967295;v, 23:59, "Ron N." <rhnlo...@yahoo.com> wrote:
>
> Look up blind source separation or blind signal
> separation.  It's research topic, usually confined
> to particularly well defined sound sources, not
> something for which you' are likely find a canned
> application for your particular recording.

As explained Ron, Blind Signal Separation (BSS) needs some assumptions
on sources, e.g. independence or sparsity.

In addition, depending on your recording conditions, you could be able
to simplify the mixing model (for example, anechoic mixtures instead
of general convolutive ones).

Try to find these informations and submit your problem to the BSS
community (http://www.tsi.enst.fr/icacentral/mailinglist.html).

Matthieu
www.ast.obs-mip.fr/puigt

>Hi Dimitris,
>
>On 28 f=E9v, 23:59, "Ron N." <rhnlo...@yahoo.com> wrote:
>>
>> Look up blind source separation or blind signal
>> separation.  It's research topic, usually confined
>> to particularly well defined sound sources, not
>> something for which you' are likely find a canned
>> application for your particular recording.
>
>As explained Ron, Blind Signal Separation (BSS) needs some assumptions
>on sources, e.g. independence or sparsity.
>
>In addition, depending on your recording conditions, you could be able
>to simplify the mixing model (for example, anechoic mixtures instead
>of general convolutive ones).
>
>Try to find these informations and submit your problem to the BSS
>community (http://www.tsi.enst.fr/icacentral/mailinglist.html).
>
>Matthieu
>www.ast.obs-mip.fr/puigt
>



Hello and thanks for the response

my exact problem is that i have recordings done with a stereo mic on a dat
machine as wav files.
they were recordings of an area that included forests, sites near the sea
near an airport and many other sites.
i want to be able to separate certain sources from the recordings and not
use the eliminated sources but the rest of the recording. what i mean is
that i want to eliminate human made sounds( cars, airplanes, ships ...)
from the rest of the recording. i hope that can assist you in giving me
some help.

Dimitris

On 3 mar, 12:14, dbormpou wrote:
>
> what i mean is
> that i want to eliminate human made sounds( cars, airplanes, ships ...)
> from the rest of the recording. i hope that can assist you in giving me
> some help.
>
> Dimitris

Hi.

The best methods to separate several sources from only two mixtures
are sparsity based BSS methods. In a first step, they estimate the
mixing matrix. Then, they reconstruct the sources, in the determined
(same number of sources and mixtures) or underdetermined [1,2] (more
sources than observations) mixtures. Unfortunately, I am not sure that
you will have a good quality of separation by only using a BSS
method.

Concerning the sparsity based methods, some of them, e.g. [3,4],
suppose that in the analysis domain, at most one source is active.
This hypothesis is very difficult to satisfy when several sources
occur.

On the contrary, several authors, e.g. [5], suppose that, for each
source, there exist a tiny zone in the analysis domain where a source
occurs alone. This assumption is less restrictive than the previous
one. Note that [5] has only been proposed for the determined case, but
using [2], you can easily extend [5] to underdetermined mixtures
(which is the situation you met).

Between both above families, you have a lot of methods that have been
proposed, e.g. [6,7].

Last but not least, you have a good survey of convolutive BSS methods
in [8].

Good luck!

Matthieu
www.ast.obs-mip.fr/puigt



[1] P. Bofill, E. Monte: "Underdetermined Convoluted Source
Reconstruction Using LP and SOCP, and a Neural Approximator of the
Optimizer". ICA 2006: 569-576.
[2] R. Saab, �. Yilmaz, M. J. McKeown, R. Abugharbieh,
"Underdetermined anechoic blind source separation via l^{q}-basis-
pursuit with q < 1". IEEE Transactions on Signal Processing, 55(8):
4004-4017, August 2007.
[3] O. Yilmaz, S. Rickard: "Blind Separation of Speech Mixtures via
Time-Frequency Masking". IEEE Transactions on Signal Processing, 52(7):
1830-1847, July 2004.
[4] P. Bofill: "Underdetermined blind separation of delayed sound
sources in the frequency domain". Neurocomputing, 16(6):627-641, 2003.
[5] M. Puigt, Y. Deville: "Time-frequency ratio-based blind separation
methods for attenuated and time-delayed sources". Mechanical System
and Signal Processing, 19(6):1348-1379, November 2005.
[6] A. Blin, S. Araki, S. Makino, "Underdetermined blind separation of
convolutive mixtures of speech using time-frequency mask and mixing
matrix estimation". IEICE Transactions Fundamentals, E88--A(7):
1693-1700, July 2005.
[7] S. Arberet, R. Gribonval, F. Bimbot, "A Robust Method to Count and
Locate Audio Sources in a Stereophonic Linear Anechoic Mixture".
ICASSP 2007, vol. 3, pp. 745-748.
[8] M. S. Pedersen, J. Larsen, U. Kjems, L. C. Parra, "A Survey of
Convolutive Blind Source Separation Methods". Springer Handbook on
Speech Processing and Speech Communication, 2007.

On Mar 3, 3:14 am, "dbormpou" <dborm...@bio.auth.gr> wrote:

> Hello and thanks for the response
>
> my exact problem is that i have recordings done with a stereo mic on a dat
> machine as wav files.
> they were recordings of an area that included forests, sites near the sea
> near an airport and many other sites.
> i want to be able to separate certain sources from the recordings and not
> use the eliminated sources but the rest of the recording. what i mean is
> that i want to eliminate human made sounds( cars, airplanes, ships ...)
> from the rest of the recording. i hope that can assist you in giving me
> some help.
>
> Dimitris

There are many audio editing tools that include some useful features.
As an example, there is a freeware Audacity:
http://audacity.sourceforge.net/

I don't suggest this as the best, but as an example. Looking at its
features:
http://audacity.sourceforge.net/about/features

"Use the Drawing tool to alter individual sample points." suggests an
editing ability to remove signal components that you can identify as
separated in time or frequency from components of interest. Audacity
provides the displays for doing this.

"Remove static, hiss, hum, or other constant background noises."
suggests more tools to remove interferences. Hum removal tools
sometimes involve adaptive filters that can identify continuing
components and remove them even when they overlap signals of interest.

Using such tools requires that the editor understand some of the
characteristics of the desired and undesired signals and uses that
knowledge to make editing decisions. Examples of this would be tones
that you can identify as sourced from man-made sources. Airplanes
often fly with constant engine rotational rates. This produces a set
of related tones and these tones can be identified and removed. There
are also click or pop removal tools for removing brief high intensity
events.

Try Audacity or go to Google for more alternatives. There a probably
better tools at greater cost.

Good luck.

Dale B. Dalrymple