Filtering/extracting vocals out of a track?

[sapito]

AFAIK its pretty hard to do a DIY job but maybe some more knowledgeable people would know more?

Been playing around with ableton lately and cant find vocal tracks for most of the samples/songs that interest me.

[Rabaelthazar]

Do you want to make us of the vocals or everything but the vocals?

If the latter, you can use a "centre cancel" algorithm which basically removes the middle part of the stereo spectrum (where the vocals are usually placed). You may be left with reverb tails in the rest of the stereo field though, so not a perfect solution but probably the best and easiest way to do it.

http://wiki.audacityteam.org/index.php?title=Vocal_Removal

If you want to make use of the vocals, it is tough. Creative eq'ing (making use of the high and low pass shelving filters) is your best option and can get you pretty close, but because a vocal performance covers a wide frequency range and so do all the other instruments in the track you'll always be sacrificing something - either the timbre of the vocal will be compromised or you'll have lots of other sound spilling on to the track.

No magical solution as far as I'm aware. Good luck.

Edited June 30, 2011 by Rabaelthazar

[sapito]

yeah sorry I want to keep and use the vocal track. Dont know how to word it properly.

I have been playing around with EQing but as you says its pretty hard to nail it.

Thanks for the help

Im curious how or where non-well known music makers get their samples without getting the raw vocal tracks? Or maybe they just use whatever is available and have better skills

[theuserformallyknownasd00d]

ive always wondered this sapito! after years of semi mastering all the other aspects, i think its a very creative and dedicated person who masters it, and i aint one!

[ballzac]

If there are sections of the track that are identical except without the vocals, you can subtract that from the combined part with the vocals:

(Vocals + instrumental) - instrumental = vocals

One complication is that most music is going to have some kind of equalisation or compression that is different when the vocals are present, to make the vocals stand out, so you will get artefacts, perhaps a quiet version of the background music, and then you can use eq and compression to try to diminish it even more. I'm struggling to remember, but I've done this effectively before, and I have a feeling I just used Goldwave. There's a noise reduction tool that allows you to load an audio file of a section of noise. If you load the background music instead of noise, it will subtract this music from the total.

Edited June 30, 2011 by ballzac

[ballzac]

Here's an example. Took about thirty seconds. A lot of tweaking of the settings is required to get rid of as much background as possible without removing too much of the vocals. This sounds pretty bad, and the beat wasn't removed, but it's just a quick example of what can be done:

atmosphere3.mp3atmosphere2.mp3

Another alternative, if you don't have your heart set on a particular track, is to make your way over to ccmixter and use some of the a cappellas there. There are plenty of really professional sounding vocal tracks under a creative commons licence.

Some of my tracks where I've used vocals from ccmixter members:

 

 

http://ccmixter.org/files/Ballzac/19565

atmosphere2.mp3

atmosphere3.mp3

atmosphere2.mp3

atmosphere3.mp3

Edited June 30, 2011 by ballzac

[ballzac]

Just another example. This time with a commercial song. The major artefacts in this one are due to the 'instrumental' part that I subtracted from the vocal part having a slightly different synth line.

It takes a fair bit of work to find parts that match precisely (and sometimes you won't be able to find them). I managed to remove the beat this time, but there is significant phasing too. My guess is that this is due to Gibbs phenomena. Goldwave is a freeware program, and I wouldn't be surprised if there are commercial programs that do a better job. Once you've used the noise reduction tool to remove as much of the beat as you can while minimising artefacts, you can use compression to make the quiet bits even quieter, so if your subtraction is somewhat effective, you can improve it in this way. Then you can use EQ to remove frequencies that don't interfere too much with the vocals. The gaps where there are no vocals can be made silent. EQing, reverb and compression in your final mix can further improve the sound of the vocals and how well they sit in the mix. If there is phasing in the final vocals, it might still work well with your mix, or it may be an annoyance, depending on the style of music you are trying to mix. As I said, there's work involved, and I don't know of a program that does this for you. It would probably take me a week to extract vocals from an entire 3 minute song to a standard that I felt was okay to use in a mix, and it might be impossible if there is, for example, a beat during the vocal parts that isn't in any other part. Goldwave probably isn't the ideal tool for the job, but I think it's great for a freeware program.

01 - Better off Alone - Alice Deejay2.mp301 - Better off Alone - Alice Deejay3.mp3

01 - Better off Alone - Alice Deejay2.mp3

01 - Better off Alone - Alice Deejay3.mp3

01 - Better off Alone - Alice Deejay2.mp3

01 - Better off Alone - Alice Deejay3.mp3

Edited June 30, 2011 by ballzac

[ballzac]

I don't know how successful the centre cancelling algorithms that Rabaelthazar suggested are, but perhaps you could use that to remove the vocals and then subtract the result from the track. If there's reverb in the vocals, that don't get removed in the centre cancelling, it might actually be a good thing because you will get a drier result when you extract the vocals, and then you have more control over the final wet mix. There may be some other stuff that is removed during this process, and this will appear in you final vocal track if you use this method, so it may not be ideal, but might be worth a try. Luckily, heavily produced music, like pop and a lot of electronica, relies heavily on placing elements in different locations both spacially and spectrally. If you're planning on using the sex pistols' vocals or something, it might be a bit less successful.

Edited June 30, 2011 by ballzac

[Undergrounder]

I don't know how successful the centre cancelling algorithms that Rabaelthazar suggested are, but perhaps you could use that to remove the vocals and then subtract the result from the track. If there's reverb in the vocals, that don't get removed in the centre cancelling, it might actually be a good thing because you will get a drier result when you extract the vocals, and then you have more control over the final wet mix. There may be some other stuff that is removed during this process, and this will appear in you final vocal track if you use this method, so it may not be ideal, but might be worth a try. Luckily, heavily produced music, like pop and a lot of electronica, relies heavily on placing elements in different locations both spacially and spectrally. If you're planning on using the sex pistols' vocals or something, it might be a bit less successful.

 

Great info thanks ballz.. i'll go through it all in more detail later..

Concerning the bolded bit:

I understand that the same sounds at the same frequencies clash and muddy things, but spacially? Is it the same reason? ie: can you fit in two similar bass lines by separating them across the stereo channels?

I had a bit of a problem recently with distortion on some of my instruments clashing... specifically I have a big saw-waved lead that has a little distortion to it, and it clashes with the other elements of the lead. So if I separated them out spatially would that reduce the clash?

[ballzac]

Yep, that's exactly it. And don't forget that the space you have to work with is two dimensional, not one dimensional. You can create an illusion of depth by applying a gentle lowpass filter (technically this is a spectral difference, but if it's subtle it gives the illusion of space) and careful use of reverb, and you can spread a sound out by using chorus and phasers. So don't feel that you're restricted to simply moving one of them to the left channel and the other to the right. One example of what can be done to give, say, a lead synth and vocals equal precedence in the music but separate them, is to apply a phaser to the synth so it spreads out and the vocals sit in the middle, and maybe apply a reverb with small size to the vocals and reverb with large size to the synth, so the vocals are sitting in a 'box' in the middle while the synth fills the space around it.

[Rabaelthazar]

I don't know how successful the centre cancelling algorithms that Rabaelthazar suggested are, but perhaps you could use that to remove the vocals and then subtract the result from the track. If there's reverb in the vocals, that don't get removed in the centre cancelling

 

I was going to mention this in my first post, but I thought it was getting a bit to technical. The centre cancel algorithms are actually based on the principal you're talking about, Ballzac, I think by adding an phase inverted left channel to the right channel, essentially eliminating anything that is identical in both channels (ie: the centre of the stereo field).

The way that what we're talking about works is that you can take a copy of an audio file, invert the phase of it and combine it with the original to end up with nothing.

That is: if you look at a visual representation of an audio file, you have what looks like a distorded sine wave with peaks and troughs either side of a zero point. Inverting the phase esentially flips the peaks and troughs so that you have a mirror image (vertically, around the zero point).

If you play the phase inverted file at the same time as the non-inverted file, the peaks and torughs cancel each other out giving you silence.

So, yes, you could use a centre cancel algorithm (or perform the process manually) and then phase invert the result and combine with the original version with vocals to get a cente only version - but!!! bass and kick are commonly panned dead centre as well as vocals so you'd end up with that in your final result as well.

Edited July 1, 2011 by Rabaelthazar

[sapito]

Thanks heaps for the detailed info guys. Im am new to this stuff some some of it has gone over my head for the time being, but will try to get a better understanding of it.

I had a quick go of goldwave last night and couldnt even work out how to add a section of the song in the noise reduction filter :s Might give it another stab over the weekend.

[ballzac]

You just select the section of the song that you want to remove, copy it to the clipboard, and then select the radio button "use clipboard" on the left hand side of the noise reduction dialogue box.

[Undergrounder]

Yep, that's exactly it. And don't forget that the space you have to work with is two dimensional, not one dimensional. You can create an illusion of depth by applying a gentle lowpass filter (technically this is a spectral difference, but if it's subtle it gives the illusion of space) and careful use of reverb, and you can spread a sound out by using chorus and phasers. So don't feel that you're restricted to simply moving one of them to the left channel and the other to the right. One example of what can be done to give, say, a lead synth and vocals equal precedence in the music but separate them, is to apply a phaser to the synth so it spreads out and the vocals sit in the middle, and maybe apply a reverb with small size to the vocals and reverb with large size to the synth, so the vocals are sitting in a 'box' in the middle while the synth fills the space around it.

 

You're a genius... I will try that. I'm just starting to realise how really important good mixing and mastering is.

Your description of the soundstage in these terms has me inspired to really get into it more. I really want my tracks to sound alive. I invested in a decent sound card and good headphones and I really want to try and get good quality, clarity and depth of sound. I think i'm going to treat myself to a raid on Amazon and pick up some good books on EDM mixing and mastering and stuff.

[ballzac]

It took me ages to realise how important this stuff is in terms of presentation of the music. I would create a piece of music that I thought was really great, but no one wanted to listen to it because it sounded boring to them. The detail and the personality that I was hearing in it because I had created it, wasn't being communicated to other people. At some stage I created some stuff that I would say is technically inferior in terms of melody, progression, inventiveness etc., but was playing around with all these EQing and mastering techniques, and suddenly people thought it sounded great. So now I make an effort to learn more about these techniques everytime I create a piece of music. It's one thing to create music for yourself and enjoy it, but music is also about communicating a mood, atmosphere, or some other sort of emotion or 'message' to other listeners, and there's not much point in it if it's not presented in a way that will allow others to absorb it.

[Rabaelthazar]

I think i'm going to treat myself to a raid on Amazon and pick up some good books on EDM mixing and mastering and stuff.

 

This is a great book and it's by an Aussie. Talks more about live recording, but really inspirational to read. I highly recommend it.

http://www.mixingwithyourmind.com/

[Undergrounder]

Very interesting (and slightly weird) table of contents there by the looks of it! That does look an interesting read, but I think I'll try something a little more traditional/technical to begin with. I lucked out on a couple of .pdfs of music production/mixing so i'll start there. 'Dance Music Manual Second Edition' and some others, if you want them I can link.

But yeah I keep having to force myself to come back to 'what am I trying to achieve here' or 'what feeling am I trying to create' in order to progress a track. You can spend hours lost tweaking and changing sounds if you're not careful. In this latest track I got up to version 50 before I had a listen back through all the versions and realised I had it at version 27 and had just proceeded to ruin it from there.

Taking breaks, listening to other music, keeping inspiration/motivation going... all so important as well.

[monkey]

Some great posts here.

As a last resort, if you can't cancel the vocals using more sophisticated methods, try cutting the frequencies 75-1100 Hz using a graphic EQ or something. Narrow down within that range using trial and error... works good enough for mashups and stuff.