on taking a track ie. voice back out of a mix...

[JeffB]

I bet if you were to split those into two mono track, panned center, and invert one of them, then mix them together they'd cancel each other out.

I think I had read before that if one can take a sound and play an identical sound with an inverted wave they will cancel each other out and there will be silence. Is this what you were referring to? It sounded pretty awesome, like someone could take out the roar of a jet or a jackhammer if they had something that could generate an inverted wave of the load sound.

Would it be possible using that principal to take out a track that had been mixed into another track if a copy of the added track were still available?

For instance, if someone had mixed in a track with someone singing and there was no longer a copy of the original music track, but there was one of the original of the singer(s), could the track with the singing be "panned center, inverted and then mixed in with the track that had already been mixed", in effect taking out the track of the singing?

[kozikowski]

Before we get all throw-uppy over this, it only works when the rest of the orchestra is in full stereo and not in the center. You can't take a mono track, split it, and delete the singer. The whole orchestration, singer and all, will vanish.

<<<inverted and then mixed in with the track that had already been mixed", in effect taking out the track of the singing?>>>

That one's a little more interesting. This would have been a nightmare in analog land, but it's possible in digital. Assuming........the voice track was mixed into the other track "straight." If there was echo or other spacial effects mixed in with it, you're dead.

Now you're going to run into Audacity's limited ability to select stuff. I could probably handle this in another editor...but even then..

My next request is for you to post a short clip of voice and the associated mix--say ten seconds of both, so we can download it and mess with it.


Oh, wait. I can think of another reason this would go into the toilet. If either track is MP3, the chances of an accurate singer cancellation are zero. MP3 works by adding distortion, the higher compression, the worse distortion. So the MP3 singer on the final mix will not match the MP3 stand-alone singer.

Koz

[alatham]

It's certainly possible, to an extent.

I remember a post on the old forum boards where someone had recorded an hour long lecture onto two separate tracks (due to an intermission in the middle). The only problem was that he had recorded the second track using the Stereo Mix source and had two tracks like this:
1) first half of lecture
2) first half of lecture + second half of lecture

After some advising, he was able to invert the first track and subtract the first half of the lecture from the 2nd track. So he was able to salvage that project doing this.

But this is a limited case. It works because there were no effects involved, and he knew exactly where the track was panned, and what the gain setting was.

So, this can work, if (as I just said) you know exactly where the sound is panned, you know exactly how loud it should be, and any effects applied were applied only to that track (you also need to know the setting of any effects used).

Experimentation can be used to find the panning and gain settings (though it'll take awhile).

Now, about the effects. If, for instance, there was reverb applied to the voice track separately and you know the exact settings used, you can still do this. But if the voice was mixed into the final track and then reverb was applied, you're out of luck.

I disagree with Koz about using mp3s to do this. It can never be 100% reliable, but high quality mp3s will be close enough (provided you can adjust for any time delay added by the encoder).

I've actually done it myself with one of the tracks on my new record. The original plan was to subtract the original working wav file from an mp3 of the same track and then use that distortion as a sound effect by fading it into the original wav file. When I first tried this, I was using 256 kbps mp3s and the mp3 was so close to the original wav file that when I subtracted them I got nothing but a very small noise floor. I had to reduce the quality of the mp3 all the way down to 96 kbps before the mp3 was different enough to leave some distortion behind.

If you're interested in knowing what the difference between an mp3 copy of a song and a wav file is, you can try this yourself. Take a wav file, export it to mp3, then re-import it, line it up (the timing will be off), then invert one track and mix them together. Anything left over is the difference between a wav file and an mp3. And it sounds awesome.

[JeffB]

...<<<inverted and then mixed in with the track that had already been mixed", in effect taking out the track of the singing?>>>

That one's a little more interesting. This would have been a nightmare in analog land, but it's possible in digital. Assuming........the voice track was mixed into the other track "straight." If there was echo or other spacial effects mixed in with it, you're dead.

Now you're going to run into Audacity's limited ability to select stuff. I could probably handle this in another editor...but even then..

My next request is for you to post a short clip of voice and the associated mix--say ten seconds of both, so we can download it and mess with it. ...

Koz

OK, I imported a little .wav file of a car crash.
I then recorded a voice track saying "Look out!"
I then exported the two as a .wav file. - The little 5 second clip includes both tracks blended together.

I uploaded the files to esnips.com an online file sharing service so that you could see if it works. I'm not sure how to "invert" an audio track.

The audacity file contains both original tracks, the clip of the car crash and of the voice over.
The .wav file has both blended together.

If I understand this correctly, one ought to be able to import the blended .wav file, add an inverted copy of the voice over which should cancel out the voice over leaving an audio file that sounds like the original before the blending.

Here is a link to the folder with the files which are online and can be downloaded:

http://www.esnips.com/webfolder/158c54f ... 6f71f7c781
or try:
http://www.esnips.com/fm/cd95cd89-cde2- ... &source=ws

Unfortunately, esnips doesn't have the ability to set up sub-folders which Audacity uses for its files. These files are all in one folder. They can be arranged in the original order pretty easily however with the Windows Explorer.

Here is a screenshot of the original folder structure:



These two files were in Audacity's data folder (labelled : test.car.voice.split_data

b00008.au
b00015.au

-------------------

track one = imported car crash wave file = b00015.au
track two = recorded voice-over = b00008.au

track 1 + track 2 = blended wave file = test.car.voice.split.wav

if we were to name:
the blended wave file track 3 and
inverted voice over track 4

a rough equation might be:

track 3 + track 4 = track 1

In effect the voice over would be "erased" from the blended track.

[JeffB]

Well I'll be darned, it worked!!!

I don't know if it's my browser, or esnips, but it changes the extension on those files when I download them. After downloading, they have to be manually renamed to .au .aup etc.

Anyway, I opened the wave file and then imported the sound track for the voice over.

I inverted the voice over track and roughly lined it up with the blended wave file, but the experiment didn't work at first.

I zoomed in and tried to align them more closely, but it still didn't work.

The I dragged down the upper & lower boundaries of each track stretching them out to fill the whole screen and tried aligning them perfectly.

But still no luck.

I zoomed in even further to no avail.

I eventually figured out I could do a selection that would span both tracks to give me a line I could use as a reference point to align them more accurately. Then I zoomed waaaay in on that end point of the selection which had been aligned with two waves that almost met in the middle. I then aligned those two waves up perfectly(?) or at least well enough that when I played it back the voice was gone! Ta da!

Now if Steven Spielberg ever calls me up and tells me that he had digitally saved a movie track and someone inadvertantly copied another sound file on top of it and he didn't have a backup, could I please fix it for him if he paid me $3 million I could say "Sure thing".

I won't hold my breath, though. He'd probably just make the poor sucker who made the mistake come here to find help on how to fix it.

Anyway, it's kind of a neat trick even if it can only work in limited circumstances.

BTW, it seems to me that it wouldn't matter if extra effects etc. had been added to the track that had been mixed in & needed to be removed later, as long as someone had an exact duplicate of whatever needed to be removed.

[alatham]

Yeah, making sure the two sounds are exactly lined up is important. You really need to be accurate down to each sample. If you're more than a sample or two off, you'll just get some comb filtering, or if it's further off, it'll sound like a very short slap-back echo.

But once you get it perfectly lined up, it's all good.

[JeffB]

...But once you get it perfectly lined up, it's all good.

To me it's like magic!

[alatham]

Nah, it's just simple matrix mathematics.

Well, as simple as adding and subtracting matricies can get.