Page 1 of 2

Ability to remove music leaving just vocals

Posted: Mon Nov 14, 2005 4:01 pm
by IanA
I'd really like to see the ability to remove music as much as possible, leaving ideally just vocals present.

I have had some success with this using the dsp_centrecut plug in, in conjunction with the user defined 'noise reduction' from the clipboard.

It'd be really good if there was any way of improving the results further.

Posted: Tue Nov 15, 2005 9:08 am
by Stiiv
Hi Ian.

There are many posts here regarding attempts to 1)remove the music, leaving the vocals, & 2)vice versa. Look around.

Short version: Both goals are just about impossible to accomplish in an acceptable fashion, Goldwave or no Goldwave. :wink:

Good luck, & enjoy the forums.

Posted: Tue Nov 15, 2005 5:03 pm
by IanA
Thanks Stiiv,

I'd already had a look at some of the posts, and have since gone right through the older posts.
Most posts really seem to be on removing vocals though rather than separating/isolating them which is what I'm trying to do.

I've also spent time looking on various sites in the past to find different ways of doing this.

I do realise it is just about an impossible task to achieve, and have tried a few different programs with various results.
Audio Cleaning Lab did give some degree of success as it allowed inverting phase of Channels and filtering out from a noiseprint.

I have found that Goldwave is much more effective though, and I've purchased a licence. :D
It'd just be nice if there were any other techniques to improve the results further.

Thanks for your reply, and I'll keep an eye on the forums for other ideas.

Posted: Tue Nov 15, 2005 6:30 pm
by Stiiv
I have found that Goldwave is much more effective though, and I've purchased a licence.
GJ, pal. A wise purchase indeed.
Thanks for your reply, and I'll keep an eye on the forums for other ideas.
My pleasure. Wish I'd had something better to tell you.

Posted: Thu Jan 19, 2006 3:05 pm
by Perticelli
it would seem to me that voice characterizatiuons and music characterizations, as represented in a wave file/frequency file signature are very different in nature.
Why cant a computer identify the subtle, yet real differences between musical sounds and frequencies and vocal sounds and frequencies?
This would allow for a program such as GW to scan and isolate the sound differences, then sparate them digitally...

is this a dead end science that's beeen tried or does it only exist on million dollar mixing bopards/computer set-ups?

Posted: Thu Jan 19, 2006 10:26 pm
by piano nick
"Why can't a computer identify the subtle, yet real differences between musical sounds and frequencies and vocal sounds and frequencies?". This question has been asked so many times on so many forums, that I have lost count.

The problem in a nutshell (without going too far into Fourier Series):

The wave (sound) of a pure A440 tone from a piano is identical to the wave of a pure A440 tone from a clarinet. Now you ask, "what is he talking about - they sound very different".

Well, note that I said a pure A440 tone, not all the overtones that make the sound of each instrument or voice unique. A pure A440 tone when viewed on an oscilloscope is what is called a sine wave (you may have to study some mathematics to get up to speed on what a sine wave is). It doesn't matter what makes the sound, if it is a pure A440 tone without any other harmonics, it is a sine wave, and all A440 sine waves sound the same. They all look the same on an oscilloscope.

The reason you can tell a piano A440 from a clarinet A440 is that there are literally dozens of higher overtones (sometimes called partials). Some of these overtones are even multiples of octaves; but most are what might be termed "odd" harmonics- they are essentially "out of tune". This is particularly true for the piano. When a string is struck at it's mid-length, it emits a particular set of overtones, but when it is struck at different locations along its length, the mix of the overtones changes with the striking position. Piano strings are struck by hammers, but the striking position is not a natural node of vibration - if a string is struck at 1/4 points, it will vibrate about these points called nods - where the string essentially is motionless. It will also vibrate about the ends (the ends obviously being stationery or fixed. The relative hardness/softness of piano hammers also affects the tone or the balance of the harmonics.

The French mathematician Fourier, formulated his ideas about sine waves: Any sound can be made up of fundamental pure sine waves of different frequencies and volumes. The multiple sine waves "mix" together - and this phenomenon can be described mathematically - I did it in third year electrical engineering.

The computer can recognize these various frequency sine waves, but the A440 of a piano and the A440 of a clarinet are the same to the computer and this holds for all the multiple frequencies of one single note. The computer cannot separate the A440 piano from the A440 clarinet - this is the basic problem. Neither can it separate the "C's" from the various instruments - because it is the "mix" that makes each instrument unique.

The same problem exists with the human voice - it generates a whole range of harmonics - if it didn't, all voices would sound the same. Therefor the computer can't separate one voice from the other and even worse, it can't tell the A440 tone of the piano from the A440 tone of a singer.

People have suggested that computer programs could learn to distinguish between the "overall sound" of a clarinet and a piano - and that is quite possible, but there are literally thousands of different pianos and there are more than thousands of human voices. Consequently the problem rapidly becomes unmanagable.

There have been bold attempts to overcome these problems, but to date (as far as I know, and I've been watching closely) there has been very little success.

One last point - you can probably tell a clarinet from an oboe (some people can't, but this is a matter of ear training and practice), but for a computer program to learn the nuances of literally millions of sounds reminds us how powerful a computer the human brain is.

The solution doesn't even exist on million dollar equipment. Someday maybe, but not for a long time.

http://mathworld.wolfram.com/FourierSeries.html

http://www.nst.ing.tu-bs.de/schaukasten ... n_idx.html

http://www-groups.dcs.st-and.ac.uk/~his ... urier.html

PN

Posted: Fri Jan 20, 2006 11:24 am
by Perticelli
truly an enlightening and excelllent reply.
Luckily, i understand all the discussion of octaves, overtones,vibrsations, etc..so i fully understand what you r saying.
I think its going to become a complex mathematical equation that eventually separates the human tones from instrumental tones..It seems computers do that now to a small degree with voice recognition software. Im sure the principle is very different, but from one technology ccan another be borne....
I believew the breakthrough will come when the software is aimed at a different place than the whole of the overtones and nuances of a sound signature..as a computer program may look at it.
i brelieve the distinguishing characteristic is actually going to be found from the BEGiNNING of the sound..the immediate initila sound characterisctic of a human sound vs an instrumental one.Of course, i cant identify what or why that will be the mark, but i have a feeling that looking at the absolute beginning of a sound, if magnified and stretched and broken down into component signatures, we will be able to find a distinct characteristic that is present in one and not the other.
As a matter of practicality, that will also allow for much faster and more efficient evolvement of the technology as it moves forward.

Ahhh,...another 10 yrs or so and maybe we'll get there...

Posted: Sat Jan 21, 2006 2:32 am
by piano nick
Pertecelli:

Just a small update:

There is a software sequencer I use that has a so-called "pitch to midi" conversion. It purports to take a wave file, and produce a midi file from that. In other words, it "sorts out" the instruments.

Several of us that use this software (we are located all over North America) attempted to produce a midi file from a wave file. I played and recorded a simple melody on the piano and "fed it" into the program. One important point - none of the notes over-lapped - that is, the second note did not start to play until the first note was finished, and so on. This is imporant because when two notes are sounding at the same time, it would be even more difficult for the software to separate the notes.

The result was at best laughable, at worst pathetic. Truly pathetic. In fact one of the guys suggested that this could be a good way to generate a new melody. This is because quite often, the melody of the "transcribed" midi file bore no resemblance to the original melody. One guy said, "it was a new tune - I didn't recognize it". These guys are all musicians - they can tell one tune from another.

Some of these guys also tried other software programs with similar results.

Now imagine what would happen if a dozen instruments were playing all at the same time (a piano can easily sound a few dozen notes all by itself by using the damper pedal). Absolutely hopeless.

I'm thinking there are much easier ways to get rich, than developing this software.

PN

Posted: Mon Jan 23, 2006 4:29 pm
by Perticelli
ha!
im not looking to get rich or make a penny, i just want new technology!!

i believe the first millisecond of the transmission..whether its voice or instrumental have a distinct difference that we havent pinpointed yet.
Without going into a wildly speculative speech, let's just leave it at that for now!

Re: Ability to remove music leaving just vocals

Posted: Fri Jan 27, 2006 1:57 pm
by Jossse
IanA wrote: I have had some success with this using the dsp_centrecut plug in, in conjunction with the user defined 'noise reduction' from the clipboard.
hey!
where can i get that "dsp_centrecut" plugin???
does it work??
:D

Re: Ability to remove music leaving just vocals

Posted: Fri Jan 27, 2006 5:47 pm
by GoldWave Inc.
You can find some information about the dsp_centercut plug-in here.

Chris

Re: Ability to remove music leaving just vocals

Posted: Fri Jan 27, 2006 11:39 pm
by Jossse
GoldWave Inc. wrote:You can find some information about the dsp_centercut plug-in here.

Chris
yes... i found the GWAmp Plugin... i downloaded it... im using it...
but the dsp_centercut???????
i couldnt install that... i thought it was an exe file!!
but in the zip folder (dsp_centercut-1_2_0.zip) there were 14 files...
i tried almost everything...
can u help me with this? 8)

Re: Ability to remove music leaving just vocals

Posted: Mon Jan 30, 2006 8:08 pm
by GoldWave Inc.
You'll need to extract the files from the ZIP archive. One of the files is "dsp_centercut.dll". Make note of the folder where you extracted it. You'd use the Options | Plug-in | GWAmp command in GoldWave to set the folder where you extracted the file. Once you set the folder, the "Center Cut" effects should be listed. You can then use them in GoldWave under the Effect | GWAmp | GWAmp command.

Chris

Posted: Tue Jan 31, 2006 1:22 am
by Jossse
ok thx man...
i copied the extracted files on winamp/plugins folder!
and i have all the effects!
:lol:

A Cappella Generators

Posted: Tue Mar 28, 2006 8:03 am
by jdeligiannis
There are two methods that I am aware of which can generate an a cappella.

1) Inverting the instrumental - If you have an instrumental of a song, maybe from a CD single, it is possible, but not probable, that you can make an a cappella out of the main mix. If the two tracks are synchronized, simply invert one of the track's waveform and mix it into the other. Everything that is identical in the two tracks will cancel-out, and the only thing that is identical is the beat. This is what the "remove vocals" feature does. It inverts the right channel and mixes it into the left.

The problems with this technique is that the instrumental needs to be available and most of the time the two tracks will have alternate speeds, and will therefore be unable to maintain synchronization.

2) CPU Generators - There are people who are using some software that is unknown to me that generates an a cappella out of a song. It cancels out the instruments and leaves only the vocals. The overall quality is reduced, and I don't think it can be done to any song, but it can be done, and some of the ones I have heard sound excellent. While writing this, I just ran across a site that is creating many of them. Check out this place:

http://www.2pacremix.com/tupac/f149-2pa ... ellas.html

Keep in mind that many of these guys are starting off with low quality mp3 files and then save the completed work as a low quality mp3, reducing quality further. If you want to keep the highest quality in your work, always keep your files as wav. Changing an mp3 back to a wav will not improve the quality. Once an mp3, always an mp3.

Post your comments. This should be interesting.