Extracting the Center Channel - possible in Audacity?


As shown in the above video. Does Audacity have anything similar? Yes, it sounds terrible, but it is fun to fool around with. It’s also possible to occasionally get something that does not sound horrible.

Tutorial - Vocal Removal and Isolation - Audacity Manual


See also the Vocal Reduction and Isolation plugin in this topic:
With a little bit of luck, it will make it into the next Audacity release, provided Gale and others can agree on it.


Robert, if you would like your tool to replace the current Vocal Remover, can you raise this on the -devel list (I “think” that is the correct list for a new feature). This would have to be agreed by 25th May and there would have to be a commitment by then that the effect would be documented in the Manual. Generally, I am supportive. :slight_smile: I think the degree of complexity is not unreasonable.

However what (in layman’s terms) do the “Isolate (Inverted)” options do?

There are two issues I see in the interface.

  • I am not clear if High Frequency Cut applies to only Vocal Removal/Isolation.
  • I can’t read the full text of the Action Choice “Isolate Center (inverted)…”. Could we say “Remove all center”, “Isolate all center” and “Isolate all center (Inverted)”? Those will fit on Windows with space to spare so will probably fit on Linux.

Plus I think the “coffee” joke in the action message ought to be something more staid.


As I’ve explained before, the inverted types are for the application on duplicated tracks.
Silencing parts in the original track gives an acapella version.
Silencing parts in the processed track gives the original version.
Letting both tracks play at the same time gives the Karaoke version.
It is in general better to work with two tracks since the band (center) can afterwards be manipulated:

  • Gain can bring back some center while suppressing the immediate neighbourhood.
  • Pan allows fine adjustment of the center position.
    But that’s all stuff for the documentation, I think.

The high cut is of course also intended for the “Vocal” actions.
This would normally be clear through grouping–hopefully available in Nyq Version 5.
I left the “Title” out for the second control because it is a pain if you’re working with a screen reader.

“All Center” sounds a bit queer to me, entire or whole would probably be better (if it wasn’t so long)
Do you have the text on two lines? I thought 15 characters would be the limit anyway.
You can try different wordings on your own, if you like, you’re a better judge.

Do you like this layout better:
“Action Catalog”
→ Analyze
→ Remove
→ Isolate
→ Isolate and Invert


“Limit Frequency Range:”
→ No
→ As given below (or “Custom”)
→ Spectral Selection
(and possible presets, Male Vocals, Female Vocals)

The two Frequency sliders

We can also make the invert option a separate control with more place to describe its purpose (less convenient though).


Plus I think the “coffee” joke in the action message ought to be something more staid.


I had to look that one up…
“Something that is staid is dignified, respectable — possibly even boring, like a staid dinner party that is heavy on the important guests but light on the laughs.”

It would certainly be more interesting to surprise the user with alternating messages.
Even the most hillarius joke gets stale over time, so why not start with a dull description in the first place?
Any proposals?
“Processing Channels”, “Extracting Stereo Information”, “Applying Action” …

I would have submitted the code already to the Devel mailing list.
However, the Nyquist version has changed to 3.09 and I had to modify the code accordingly.
Not easy since the recent GUI changes have hampered proper testing.
Some things are still not clear e.g. if I have to take the input validation into account (see post in the Quality section of the mailing list).
That’s why I’ve written “with a bit luck” :wink:


I can hear that the “Inverted” options give extra flexibility.

But on a track with well centred vocals it seems to take a lot of work to make removal/isolation on duplicated tracks sound “better” than using “Remove Vocals” or “Isolate Vocals” on one track. Does inversion become more important on a more “difficult” track?

The rub is that the “Inverted” options are harder for users to understand so need careful documentation. If the documentation is not adequate this is a strong reason not to release an effect.

OTOH I did something perverse and isolated vocals (not inverted) on a duplicate track to strengthen a weak vocal line. That retained more of the instrumentals than applying negligible vocal isolation to the original track. So strengthening the vocals or instrumentals could be mentioned as something else that could be done with the duplicate track method.

If we keep the existing layout, I would find the following order of actions better, putting the “advanced” inverted options at the end and dropping the “Catalog” word:

;control cat "Action" choice "Analyse,Remove Vocals,Isolate Vocals,Remove Center,Isolate Center,Isolate Vocals (inverted),Isolate Center (inverted)" 1

I think we could forget having “Full” or “All” in the action names. The control “Low Frequency Cut in Hz” makes clear that Full Range applies to all the options except “Remove Vocals”, “Isolate Vocals”
and “Isolate Vocals (inverted)”.

“Low Frequency Cut (Hz):” with the colon after “(Hz)”, and similarly for High Frequency is closer to our standard.

Does Strength apply to all the options except Analyze? If so, how about renaming that to the shorter “Effect Strength”.

I certainly like grouping the inverted options together. I don’t see where you are specifying Vocals or Center to act on.

I don’t know what the “Spectral Selection” choice does as compared to what “Custom” does.

The Male/Female presets may help less able users.

What would the default option in “Limit Frequency Range” be? Does being able to act on the frequency range of the Center and having presets offer enough advantage to be worth the extra control?

Probably not worth the extra inconvenience. If someone uses an inverted option in error on a single track, it should not sound very different to the uninverted option.

Thanks, Robert.


I like “Processing ”.

“Processing stereo vocals…” perhaps? Up to you.

Another minor detail. We decided to become “modern” so the error message:

This plug-in works only with stereo tracks.

should not have the hyphen in “plug-in”.


Yes and no, it is just another method, aimed to non-destructive work.
Another plus is that you can render the two tracks into one and save the time for a separate application of the plugin.

Certainly, that’s one of the highlights. It is especially interesting when amplifying dialogs in films or “Cocktail party” talk.

A more advanced example is here, where the lead is doubled, panned and the secondary voice totally modified:

Most of your proposals are now in version 1.52:
vocalrediso.ny (10.6 KB)
Main changes:

  • Overall layout the same
  • “Action” instead of “Action Catalog”
  • default index = “Analyse”
  • order not changed (I like it better to have 3 similar actions together)
  • inverted types re-formulated to “Isolate Center and Invert”
  • filters are accordingly e.g. “Low-cut for Vocal Range (Hz)”
  • author and description in about menu
  • “Strength” instead of the long version (in the 2D Stereo Toolkit, it made sense)
  • The analysis shows now also the physical pan position (First entry).

This seems extremely helpful for difficult tracks where the vocals are not in the middle.
It works like this:
You select a short portion of the vocals that you want to remove or isolate, where no other instruments are audible.
Run the analysis.
Write down the value for the pan position and enter it with reversed sign into the Pan dialog of Audacity, i.e. 0.42 gets -0.42
Render the whole track, the target band is now exactly in the center (if you’ve chosen a proper passage).
Make your desired action.
You can bring back the track into the old pan position, if you like-- you probably have to amplify it as well.
However, it’s a one-way ticket, you can’t remove multiple bands like that at the same time.
Nevertheless, it works great, especially if the vocals are not too far off the center.
Here’s a sample file that you can play with:


Thanks, Robert. I’ve attached a version corrected for “plugin” without the hyphen.

I like the text changes and am happy with the action order now that you use the “and Invert” syntax.

I doubt Analyze should be the default action - naive users will be confused by what happens, and how many pop tracks actually have any vocals on their own? Also the analysis is only for the uppermost selected track as far as I can see.

Fortunately the user’s removal or isolation choice will be saved so it will not be a problem in a new Audacity session.

I do see the considerable advantage of the pan reporting, but it adds to the documentation burden. I hope we have a volunteer to document this effect - do we?

Is it possible to add a sentence to the message Robert, something like the following where we would reverse the sign of the pan value?

Pan position: -0.15248. You can apply +0.15248 to the track's pan slider to center the pan, which may assist removal or isolation.

Longer term, Nyquist 5 (?) could the plugin detect the pan and have a control to correct it?

vocalrediso.ny (10.6 KB)

I forgot to correct it to the new nomenclature.
I try to write it connected for the most time, although I hate that my screen reader says “Ploodjin” :wink:

Analysis has the default index because vi-users would discover the menu point only by accident.
It’s a combo box and a listbox would work better since it announces “2 of 7” for the second entry, this by the way.

Former versions of Nyquist plugins (1, 2, 3) have opened a separate message box for each selected track if a string had been returned.
The idea is to show “Help” only once.
I’ve never supported this restriction, it defeats some plug-in functionality.
Compare e.g. “Replay Gain”
The solution would be to make it an analyse plugin.

And place it at the end of the screen?
I don’t know if the sentence isn’t too long for some screens, probably not.
I want to display the stereo width as well.
A value near 0 % would indicate that a single band is selected–good for centering this band and removing it.
It is not a point that demands explicit documentation.
I think we could append it to the actual plugin description as an advanced example workflow.
This hasn’t to be available immediately, has it?

Don’t know yet, I hope so.