How to remove background audio from dialogue?

[helogg]

I have a recording which consists of someone talking over a Youtube video, so there are 2 elements to this one sound file:

1. a microphone recording

2. the audio from a YouTube video (going directly through the computer, not out of speakers and through another microphone)

These two pieces of sound are in the same stereo audio file. Is there a way to lessen the sound of the audio from the YouTube video, so that the microphone recording is easier to hear? If I have an exact copy of the audio from the YouTube video, is there a tool I can use to combine that with the file I have now to achieve the result I want? I'm thinking of how noise removal tools can use a noise profile, but this would be slightly different. It would be useful if the tool could automatically detect when the audio from the YouTube video starts and stuff so I don't have to try to line anything up manually. Does such a tool exist? I feel like this is probably a common problem.

Thanks in advance for your help!

[Drew Stephenson]

Depending on the nature of the youtube audio, if you have a precise copy then you should be able to flip the polarity. Align the two pieces and you should be left with much clearer dialogue.

I'm sure there are tools that will help with the alignment but I don't think any of them are cheap or designed for this specific job - so you'd be experimenting.

Smarter people will probably be along shortly with better advice.

[helogg]

thanks for your response, my only idea was to switch the polarity in a DAW and try lining it up and exporting it but the problem would be lining things up, but i wasn’t sure that works or how that might sound different versus if the removed sound have never been there in the first place.

are there any tools out there that can automatically detect when a given sound starts or stops in another sound, and removes it? kind of like how some denoisers can be fed a sound profile?

thanks again for your help!

[Mike Stranks]

My first thought was the 'polarity flip' - and by all means give it a try...

... but I think you'll almost inevitably severely compromise the audio you wish to keep, possibly to the point where it's unusable...

Some of the tools available now are very good, but even if you pay megabucks can only do so much. The old analogy of getting the eggs back out of cakes is still largely true.

But good luck with your efforts... I hope you achieve what you're aiming for.

[The Elf]

An expander might help, up to a point.

Aligning an opposite polarity track is not so onerous. There's no need for any device or software to align it for you - just a bit of carefully applied donkey-work.

Firstly, find a exposed piece of unwanted audio in the 'dialogue + unwanted YouTube' track. Loop around this section. Flip the polarity of the 'YouTube only' track and now begin to align the two as closely as you can. Use the waveform display and set the editing granularity lower as you get closer to perfection - down to sample accuracy in the last steps.

As you get closer to perfect alignment you will hear the combined audio reduce to a point where you can can get no further improvement. You may then need to adjust the level of the 'YT only' audio to where it again produces the best cancellation.

Hopefully this will get you a result you can work with. If you are still getting a small amount of bleed you can then combine this technique with expansion.

[Drew Stephenson]

You could then apply Izotope's voice de-noiser from the Elements version of RX. It's frequently discounted to as little as $29.

[Hugh Robjohns]

Izotope RX dialogue extractor?

[James Perrett]

I'd still say that it would be best to invert the polarity of the Youtube recording, mix it with the combined recording and adjust the level of one of them for maximum rejection. We don't know what is on the Youtube recording so, if it includes voices, most of the more fancy suggested treatments simply wouldn't work.

[Sam Spoons]

The OP has the recording with the voice added and the original FB file, and the FB audio background was recorded digitally behind the voice so the only difference between the two files is the voice (i.e. there is no speaker or room colouration to the FB audio). Given those conditions surely it should, theoretically, be possible to completely remove the FB background audio?

[James Perrett]

Given those conditions surely it should, theoretically, be possible to completely remove the FB background audio?

Provided nothing else has been done to the YT audio between the two recordings, it should be easy to remove the YT audio from the voice recording. I do things like this all the time.

[Sam Spoons]

The OP implies this is the case but was asking for a means of automating the procedure.

[helogg]

The OP implies this is the case but was asking for a means of automating the procedure.

Yeah that would be ideal, another guy has said "it should be easy", is there a particular good tutorial online for how to do this, if it cannot be automated? Thanks for your help!

[Sam Spoons]

What is your audio software and how familiar are you with it?

If you have a basic knowledge of a multitrack DAW it should be fairly easy to do, basically you'd import the two files into two tracks of the DAW, line up the background audio so that both files have the background in sync (easier to do if the background file has some easily recognisable features) and reverse the polarity of the 'before' file. Fine tune by moving it forwards and backwards until you get best cancellation. I don't know of a video showing exactly this process though but I suspect a dedicated software would take as much learning/faffing to get a good result as doing it manually.

If you're not in a hurry I'd be happy to have a go for you, assuming the files aren't sensitive.

[Kwackman]

another guy has said "it should be easy",

That sounds like the guy to ask?!

[James Perrett]

another guy has said "it should be easy",

That sounds like the guy to ask?!

That was probably me...

In Reaper you put both audio tracks on adjacent tracks. Slide one of them in time until they both line up exactly (and I do mean absolutely exactly) at the start. Go to the end and check that they still line up. If they don't you will need to change the playback rate of one of the items by hitting F2 and then entering a number in the playback rate box (just make sure that "Preserve Pitch When Changing Rate" is not ticked).

Once they are lined up, click the phase invert button on one channel (a circle with a line through it) and the unwanted audio should be reduced. To reduce the unwanted audio further you will need to adjust the level of one of the tracks so that the Youtube audio is the same level from both.

And that is basically it. If that procedure doesn't work then it is likely that one of the Youtube tracks has had different processing applied.

[The Elf]

is there a particular good tutorial online for how to do this, if it cannot be automated? Thanks for your help!

I've already explained it in detail. If there's something you don't get then let us know and we can try to clarify.