any ideas please? unmixing a track...

Hello - I’m glad the subject didn’t put you off…!

I have a stereo track which should be dialogue on Ch.1 and audio timecode (this is a shrill kind of audio signal of fairly consistent amplitude which indicates to those devices that can read it what the timecode was) on ch.2 - however the track has been recorded badly: dialogue is twin track mono on both channels and the audio timecode is on channel 2. The dialogue on channel 2 is corrupting the timecode signal so I want to remove the dialogue from track 2 to hopefully improve the signal.

Normally I would imagine this is more or less impossible - however, a clean version of the dialogue is on channel 1 so I’m wondering is there a way to use that to subtract the dialogue from channel 2?

I half thought that by moving channel 1 against channel 2 by a small number of samples and then mixing it down it would kind of phase itself out. But i"m no expert…

Thanks, tolsky.

This may be one occasion where the subtraction (inversion) method works,
If the two versions of the vocals are in perfect sync they cancel out if one is inverted and levels adujusted so they are of equal amplitude.
[ destructive interference style ]

Thanks for your reply. Can i just check on how you’d do that in Audacity then.

select channel 1
adjust amplitude if i think necessary
apply the invert filter
project menu>quick mix

i hope that’s not right cause I tried it and it didn’t work!

Thanks, tolsky

It’s not “push this button and everything is fine.” There will be one and only one critical volume of track one dialog which will cancel the dialog leakage from track two.

If somebody forced me to do this, I would probably split the tracks and use the envelope tool (two white arrows and bent blue line) to produce a dialog track that ever so slowly changed volume from maximum to zero. Then do the subtraction trick and listen critically to the results. If the dialog leakage changes character but never changes volume, then you’re dead. The two voice tracks aren’t mirror images of each other. If at one time the dialog seems to get very quiet, then use the time display to figure out which of your volume changes did the trick. Cancellation is very critical.

The leakage must be pretty bad. Most SMPTE Time Code readers can handle a certain amount of distortion and leakage. It was designed for the cue sound track on a quadruplex videotape machine which sucked on a good day. And yet people edited with it.

Sometimes you can use the volume effects tools to just make the TC track a lot louder. SMPTE TC is non return to zero square waves – intentionally – and you can hard clip them without losing any of the information. Try the Leveler tool, or apply Amplify and let it clip a bit.

It’s supposed to sound terrible. If your dog can stay in the same room with you while you’re editing SMPTE TC, then you’re doing it wrong.


If you attach a sample, a stereo WAV few seconds long*, to your next post I can try subtraction method using Audacity.
The two vocals must be in perfect sync to cancel out, if they aren’t it won’t work.

[* attachments must be less than 1Mb}

Thanks for your posts so far, I’ve attached a wav file to this one.

As for results yes I’m just after something better - the timecode reader (I’m using FCP Aux TC to generate a list) is reading it but jumping to so many false results that I’d like to improve the source if its possible.

What is the ‘subtraction trick’ exactly? or even roughly. i can’t find it at all…

Thanks again

  1. split stereo track
  2. invert left track (the vocal-only one rather than the other to maintain the correct polarity of your timecode track).
  3. Re-label both tracks “mono” (rather than “left” and “right”)
  4. Press shift and press the play button simultaneously (will play as a loop).
  5. Adjust the level (-/+ slider) of the track you inverted until the amount of vocal you hear in the sync-code track is minimised.

The good news: the two vocals in your example are in sync so the inversion subtraction method will work.

Now the bad news: the equalization of the two vocals is not the same, they need to be exactly the same for total cancellation.

In the attched file I’ve used the inversion-subtraction method without altering the equalisation, so it’s an improvement but not 100% cancellation …

You’ll have to experiment with the equalization of the vocal-only track so it exactly matches the vocal in the sync-code track to improve on my effort.

Just to finish this thread off, i had 5 files and 3 of them had readable timecode (in places) despite the overlying dialogue. The invert-subtraction method cleaned them up further but my analysis of the timecode was no better, I believe because of problems in the timecode audio.
The other 2 tapes, one of which I sent you a clip of, was simply hopeless throughout and therefore I doubt it was the dialogue causing the problem. The timecode itself was just somehow not right and didn’t do enough of the things to make a timecode reader happy!
If anyone with an interest in timecode (rather than audio processing) wants a look at such a file let me know as I’d be keen to know if there’s a way of getting something meaningful out of such a file.

Thanks for your help in teaching me inversion subtraction though.

Have you tried amplifiying the timecode track (after subtraction) as Kozikowski suggested …

[Although your timecode looks more triangular than a square wave to me].

Maybe a narrowbandpass filter which keeps the timecode but rejects most of the audio…

Unmixing is definitely possible.

Here’s some stuff I did…

[Adverts removed]

They’re impressive samples. Are you going to expound on how you accomplished it or is it just an advert?

I’m going with no, and yes.