ScoreHero
Home | Forum | Wiki
Inbox [ Login ]Inbox [ Login ]
SearchSearch MemberlistMemberlist
ProfileProfile Log inLog in
TUT: (Sort of) separating guitar from music for customs

 
Post new topic   Reply to topic    ScoreHero Forum Index -> Software
View previous topic :: View next topic  
Author Message
HansenIsGod  





Joined: 07 Dec 2007
Posts: 46
Location: Germany

PostPosted: Thu Jan 03, 2008 9:21 am    Post subject: TUT: (Sort of) separating guitar from music for customs Reply with quote

Okay, so you want to make a cool custom song, but the artist was not kind enough to provide you with separate master tracks for the guitar and the rest of the band? Unfortunately, it's impossible in practice to perfectly split the guitar from the rest of the mix. So what you could do is call your friends and record a cover of the song, or you could use the final mix in all three streams in GH - or you use the final mix on just the guitar track. That's not that cool - if you miss a note, the song will just play on as if nothing happened, or will go completely mute.

Since GH works by mixing a 'band' track and a 'guitar' track and will mute the guitar if you miss, one has to figure out a way of making two tracks that will sound like the original if mixed together, and sound like the original without the guitar if just one of them is played.

The approach I tried here is based on using band pass and band reject filters to split out the frequency range where the guitar is mostly predominant, i.e. the mids (400-3500 Hz).

Compared to the usual "all 3 streams the same/only guitar" method, this has the following advantages and disadvantages:

- it's a chore
- of course, it will not perfectly separate stuff (because that's impossible)
- the resulting "mixed" audio might sound a bit different than the original, depending on how good your filters are
- vocals will be in the new guitar track and therefore will be also muted if you miss a note (this is unavoidable since the periodic parts of speech are in the same frequency range)
- if you miss the last note before a vocal solo, you're out of luck

+ the player will get a better reaction upon missing a note, i.e. cutting out the guitar instead of nothing happening or the song going completely mute
+ whammying will now effect only guitar (well, and vocals) instead of everything

Now that we know the nuts and bolts of this method, let's begin a step-by-step walkthrough of how to do this using the free audio editor audacity.

Open audacity and import the audio file you would like to work with. (FYI, I use Gamma Ray's 'Heaven Or Hell' here).
Since we will do some filtering and get 2 tracks that will be later played back simultaneosly, we first must reduce the volume of everything to avoid clipping (digital distortion).

To do this, go to Effect --> Amplify:


Reducing volume using -3 dB is fine.


Now, we want to create a second stereo track. Go to Tracks --> Add New --> Stereo Track


Double click the first audio track to select everything, then copy it to clipboard:


Go to the beginning of the second audio track and paste it there:


Good, now we have two copies of the mixed audio. Let's split it!

The first track is gonna be the band track, so we want to apply a band reject filter that cuts out the midrange (i.e. guitar) and keeps everything else. Select the first track and go to Effect --> Equalization.


Now you will have to draw your filter curve. Remember that we want to cut out the midrange, I used the range 440-3300 Hz, but you can tweak this a little. Filtering by 30dB is ok, going deeper will give you ripple problems with the filter. You want your transitions to be as steep as possible - smooth filter curves gave me strange flanging sounds when mixing the tracks back together. Between the transitions, take care that the curve is flat. Drag the "filter length" all the way to the right (largest size, best precision). Here's how my curve looks (you can save the curve as a preset to use again later):

Press OK and it will do the filtering.

For the guitar track we will have to apply the exact(!) opposite filter. Now this is important, since if your filter curves don't match really good, you will produce holes or spikes in the spectrum which will sound nasty.

Unfortunately, you cannot just tell audacity to invert a curve, so you will have to swap your points by dragging them while remaining at the same frequency.

Select the second track, and again open the equalization menu.
Your band reject filter should still be there. To turn it into a band pass, pull the bottom points up and the top points down - do not change their horizontal position. It should look like this:

Press OK.

That's it! Filtering done.
Now, you can test your separated tracks right away: Just press play (SPACE) and you will hear how they sound mixed together. If you MUTE the guitar track, you will hear what will happen if you miss a note ingame.


If you like what you hear, save it! If not, go back using undo and fiddle around a bit. Select 'Export Multiple' from the File menu.


Set your format to whatever you want (e.g. 128k MP3), select the place where the files will go, "Numbering consecutively", split based on tracks.

Whee! We're done.

Now you can use the guitar track as guitar and the band track on the song section (use only 2 files and put silence on the rythm/bass track, or the filtering will not be very effective).

Go and try it now!
I've made a little demo for interested, yet lazy people using "Tell the Devil" by Predator, available on Jamendo (Licensed CC-BY-NC-SA)

Here's the demo: http://www.savefile.com/files/1298823

Feel free to tell me what you think! Maybe this could be automated by i.e. using SoX (haven't toyed around with that much) if it has an appropriate filter. Of course you can also use other editing software like Audition.

On a sidenote: Avoid transcoding! It's sad that GH will just take 128k non-joint stereo MP3 (that's soo 90s), but don't make things worse by going from mp3 to mp3, as this will cost even more quality. The best thing to do here is rerip from CD. Best to use the latest LAME beta (3.98b6). If you're interested in encoding, audio quality and such, you will learn a great deal at www.hydrogenaudio.com
Back to top
View user's profile Send private message
st4t1cv01d  





Joined: 17 Sep 2007
Posts: 28
Location: Munich, Germany

PostPosted: Fri Jan 04, 2008 9:58 am    Post subject: Reply with quote

That's very cool. Thanks for the guide!
Back to top
View user's profile Send private message
RGBrewer  





Joined: 18 Oct 2007
Posts: 8

PostPosted: Wed Jan 23, 2008 1:02 am    Post subject: Reply with quote

What about then attempting to remove the vocals, thereby leaving only the band track? Ala... http://audacityteam.org/wiki/index.php?title=Vocal_Removal
Back to top
View user's profile Send private message
stereth  





Joined: 17 Jul 2007
Posts: 82

PostPosted: Wed Jan 23, 2008 4:08 am    Post subject: Re: TUT: (Sort of) separating guitar from music for customs Reply with quote

HansenIsGod wrote:
For the guitar track we will have to apply the exact(!) opposite filter. Now this is important, since if your filter curves don't match really good, you will produce holes or spikes in the spectrum which will sound nasty.

Is there a way to create a difference track? ie, band track = original track - guitar track. Or does that depend on the format? I have no idea how it would sound on its own, but if it's possible, it would match up to recreate the original track.
HansenIsGod wrote:
On a sidenote: Avoid transcoding! It's sad that GH will just take 128k non-joint stereo MP3 (that's soo 90s), but don't make things worse by going from mp3 to mp3, as this will cost even more quality. The best thing to do here is rerip from CD. Best to use the latest LAME beta (3.98b6). If you're interested in encoding, audio quality and such, you will learn a great deal at www.hydrogenaudio.com

Very true, but I'd like to point out that you may already be doing this.

Frequency-domain filters don't work on mp3s, only on waveforms. If you import an mp3 into Audacity, you have decoded it. If you export an mp3 of the filtered waveform, you have reencoded it.

The real best thing to do here is what every professional who works with audio, video, or graphics does: work uncompressed. Rip from CD as a wav file, edit as wav, export as wav, and import the wav into GHex. It will cost you some hard drive space (CD-spec wav files are 10.5mb/min), but it will prevent any quality loss, and will also save you time because there will be fewer conversions taking place.
Back to top
View user's profile Send private message
HansenIsGod  





Joined: 07 Dec 2007
Posts: 46
Location: Germany

PostPosted: Wed Jan 23, 2008 11:52 am    Post subject: Reply with quote

RGBrewer wrote:
What about then attempting to remove the vocals, thereby leaving only the band track? Ala... http://audacityteam.org/wiki/index.php?title=Vocal_Removal

This method works completely different (makes use of how tracks are mastered in regard to stereo panning), and most likely will not work for guitars. I don't see the immediate benefit in doing this for GH customs (for Rockband, maybe?), except if you would want to really split out the vocals, then do the guitar/band separation, and try to add the isolated vocals back into the band track. Might work in theory, but I doubt it'll sound nice in practice (the vocal separation with this method hardly ever works even remotely correct in real life). It also makes the process a lot more complex and time consuming, when with just the EQs, it works "well enough".
However, I might give it a try and check how the results sound.

stereth wrote:
Is there a way to create a difference track? ie, band track = original track - guitar track. Or does that depend on the format? I have no idea how it would sound on its own, but if it's possible, it would match up to recreate the original track.

You mean like, create the guitar track via this method, and then inverse mix-paste it to the original (aka subtracting the guitar track)? Another thing that might work in theory, but I haven't tried out yet in practice; I will try it later when I have access to my media rig. This method depends on phase relations, and I'm not sure how well these are preserved when doing lossy encoding of the splits. I doubt it'll sound better than the filters alone; if you have your filters working, they are quick 'n easy to use. Some other software can even "invert" the filter more easily, but I haven't found a way for this in audacity yet.

stereth wrote:
Very true, but I'd like to point out that you may already be doing this.

Frequency-domain filters don't work on mp3s, only on waveforms. If you import an mp3 into Audacity, you have decoded it. If you export an mp3 of the filtered waveform, you have reencoded it.

The real best thing to do here is what every professional who works with audio, video, or graphics does: work uncompressed. Rip from CD as a wav file, edit as wav, export as wav, and import the wav into GHex. It will cost you some hard drive space (CD-spec wav files are 10.5mb/min), but it will prevent any quality loss, and will also save you time because there will be fewer conversions taking place.

I'm aware of the fact that mp3s are decoded upon loading them in audacity, thanks nobody said I would do that. I don't work with uncompressed files, but rather losslessly compresed material (FLAC) which saves some space. (Sidenote: You can directly "EQ" a mp3 if you directly manipulate the MDCT coefficients and/or frequency bins of the encoded file. That's what cheap mobile mp3 player EQs do, and that's also why these EQs usually sound like crap).

For PC GH3 modding, you must use 128k mp3s, while when doing customs for other plattforms, you're not bound to that and it is indeed good advice to use wav as the last intermediate.

On the other hand, keeping in mind that we're doing this for a GAME (that won't even let me use my ASIO interface for playing ), transcoding doesn't matter that much, if the source material is good, usually nobody will be able to tell the difference (with recent MP3 codecs). For professional work of course, it's a No-No.
_________________
Back to top
View user's profile Send private message
Headbang  





Joined: 28 Sep 2007
Posts: 474

PostPosted: Wed Jan 23, 2008 12:40 pm    Post subject: Reply with quote

Thanks for the great guide!

I'm going to be trying this.
_________________
Justin Hawkins signed my butt.
Back to top
View user's profile Send private message
stereth  





Joined: 17 Jul 2007
Posts: 82

PostPosted: Wed Jan 23, 2008 11:42 pm    Post subject: Reply with quote

HansenIsGod wrote:
For PC GH3 modding, you must use 128k mp3s,

Yeah, I didn't know that. That changes things. And, certainly, if it sounds good, run with it.
Back to top
View user's profile Send private message
GuitarsNGirlz  





Joined: 25 Feb 2008
Posts: 9

PostPosted: Fri Aug 29, 2008 6:58 pm    Post subject: Reply with quote

Hey.. Thanks for the neat Tut.
few questions though..
What version of Audacity are you using? I have it too, but i am seeing the things like the equaliser different on mine..
Also, should i be able to hear the drums in my guitar track? or is it just the frequency that the drums are at? Because i am just trying to separate the Guitar from the main audio to help while trying to make a chart.
Thanks again.
Back to top
View user's profile Send private message
EpicKris  





Joined: 28 Dec 2008
Posts: 1

PostPosted: Sun Dec 28, 2008 2:00 am    Post subject: Rhythm / Bass Reply with quote

Hey, if I would like to do the same as this but with rhythm / bass instead, would I use the same method but with different frequencies and what would those frequencies be?
Back to top
View user's profile Send private message
Display posts from previous:   
Post new topic   Reply to topic    ScoreHero Forum Index -> Software All times are GMT
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum




Copyright © 2006-2024 ScoreHero, LLC
Terms of Use | Privacy Policy


Powered by phpBB