Background track isolation using multiple sources

Forgive me if I’ve posting this in the wrong room, it’s more of an audio question than an audacity question.

I want to know if it’s possible to take 3 or more audio sources, which all use the same background music track but with voice overs, and combine them to re-produce that background track without any other sounds in the different recordings. Obviously the more sources the better, but at the moment I’m having a hard time searching for the solution as I’m not sure what what I want is called.

I did try some standard VR&I methods, but the results aren’t exciting.

a free plugin called Kn0ck0ut can produce what is common to two tracks, (aka center-pan isolation) …

Not sure that will do what you describe, and it does create plenty of digital artifacts.

If the complete music , without voice over, is available in parts across multiple tracks you may be able to edit those voice-free parts together, but the result may be like Frankenstein’s monster.

The center pan isolation does work with some stereo sources, it’s a fascinating calculation. But for what I’m talking about, the calculation requires 3 or more inputs, so any plugin that takes more than 3 stereo sources (>=6 channels) would be an indication that it would be doing what I need.

Maybe it’s not possible. But experienced audio people would be able to say.

It’s not possible by simple addition/subtraction, but it is possible (in theory) by clever FFT analysis. Unfortunately I don’t know of any software that actually does this.

You would think you would be able to do this with simple waveform subtraction and management, but that would only work if you had first level studio tapes or recordings. The minute you hit internet downloads, the compression and wave damage is different for each track and they’re not wave-for-wave sisters of each other any more. Koz

Even with studio masters (mixed) you can’t do it. You can subtract the part that is common to each (the background) but you can’t subtract the part that is different. It’s the same argument as why centre-pan removal works but centre-pan isolation doesn’t work.

Historical noise reduction using multiple amplifiers

In the 1970s analog audio noise reduction was achieved by running a high numbers of identical microphone preamplifiers in parallel. The principle is just simple. If you give the same input signal to many identical amplifiers, then the part that is common in all amplifier output signals is the amplified audio input signal, while the part that is different in all amplifier output signals is the amplifier noise. You only need a high enough number of parallel amplifiers and add the output signals, then the signal differences (the noise) will statistically chancel out each other and the overall noise in the sum of the amplifier output signals will get lower.

But unfortunately two amplifiers in parallel do not reduce the noise by half. Because noise is a statistical phaenomenon with no phase correlation at all, two noise signals in parallel will only give a tiny bit smaller noise signal. This means that you must run minimum several hundred amplifiers in parallel to achieve an audible noise reduction effect. When the first analog integrated semiconductor circuits appeared in the 1970s and it was possible to produce several thousand transistors on the same chip this was really tried (and then given up very soon because the required efford was much bigger than the resulting effect was worth). But the important point here is that is was proven that this method really works.

Voice reduction using multiple sources

This means that if you have a possibly high number of audio sources, which all use exactly the same background music (audio signal) but with different voice overs (noise), then you must just align them with the background music exactly in parallel and mix them together, then the background music will get louder and the voices will get lower. The higher the number of audio sources, the lower the voices will become in the mix.

Unfortunately you will run into the same statistical problems like the “parallel amplifier noise reduction” from above. This means that you will probably need several hundred different audio sources with exactly the same background music and different voice overs to get a sensible “voice reduction” effect and I’m afraid that because the background music will not be exactly identical in all audio sources you will get a big statistical “sound mud” as output signal.

Plain english version: Load all audio sources you already have with “File > Import…” into Audacity, align them with the time-shift tool, play them in parallel and listen to the result. If the voices do not get at least a tiny bit lower or if the music gets out of sync, then the audio sources are not good enough for this and you will not get the desired effect, not in this or any other way. If it really works then try to find more audio sources and try to run them all in parallel.

  • edgar