[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: separating voices and music in movie scenes

Dear Christian,

Audio professionals agree that this popularly demanded task (although
typically in the form "separate vocals from background music in a
song")  is something you cannot really achieve.  The simplest method
is to mix the L and R channels of a stereo file into a mono file, with
phase (polarity) of one channel inverted. This cancels out everything
that was mixed into the centre of stereo panorama, which, in a typical
pop song, is vocal. But whichever instruments were mixed into center
will be lost too, any reverb or off-center vocal will remain there.
And the result will usually sound terrible, and will be in mono.
Furthermore, the dialog/music separation in a movie scene will
probably not follow the vocals/music rules of a pop song mix and you
may get absolutely nothing. Finally, in this way you can get rid of
the center (vocals) and keep the "sides" (music), but not vice versa.
Anyway, this method should be available in any audio editor, either
directly, or as sequence of simple operations.

I know that one commercially available software, Adobe Audition, has a
more advanced tool, which allows decreasing OR increasing amplitude of
any location in the stereo panorama relative to other sounds, the
effect may be limited to a certain frequency range, there is a number
of additional parameters to fiddle with, and the result remains in
stereo.  But the process is prone to ringing artifacts, similar to
overdone spectral subtraction noise reduction, it's still hard to get
good separations, and I was never able to get satisfactory results out
of it. I have not tried very hard however, I admit. Audition is, I
believe, available for download as 30-day fully functional trial, so
you may give it a try. If your soundtrack is in stereo, if you're
skillfull, patient and lucky, you may get what you need.

There may be better algorithms in top professional editing systems,
but the price would probably not reasonable. And as the task is not
what audio professionals typically do (they have separate tracks, so
they don't have to un-mix them from a final file), I have some doubts
if you will find any.

I am afraid that the old saying "you cannot take eggs out of an
already baked cake" still holds, and the only way to get good results
is to get hold of original separate dialogue, music and effects
tracks. Maybe studios will be willing to support science?

Oh, one more idea. Typically, you can buy separate "soundtrack" CDs.
You may replace the original audio with such soundtrack, to get a
no-dialogue (and no sound effects) version. Maybe, if you are lucky,
by subtracting the "soundtrack" from the movie audio, you could obtain
a dialogue+effects version. But again, different processing/mastering
may have been applied to the final movie audio and to separately
issued soundtrack, so the effect may be not so good. Or it may be a
complete disaster, if there was time drift between versions.

Good luck - and please let me know if you succeed in any way,


On 27/01/2008, Christian Kaernbach <auditorymail@xxxxxxxxxxxx> wrote:
> Dear List,
> For a study on the emotional impact of movie scenes with and without
> music we would love to learn of possibilities to separate the voices of
> a dialog scene from the underlying music. We want to take classical
> movie scenes from DVD or the like and produce versions without music, or
> without dialog.
> Is there any ready-made software that could us help to do this? Or a set
> of software routines that we could use? Freeware would be preferable,
> but commercial software could be considered. The quality of the result
> is more important than the price (as long as it it reasonable). If
> algorithms are available but "way too complicated for you, you naive
> consumer-type end user", we would be happy to start a cooperation.
> Any hints welcome.
> Best,
> Christian Kaernbach

Pawel Kusmierek PhD
Department of Physiology and Biophysics
Georgetown University Medical Center
The Research Building WP23
3970 Reservoir Road NW
Washington, DC 20007
phone: +1 202 687-8851 or 8028, fax: +1 202 687-0617