Does anyone knows if there is some tool to reverse an audio wave?

Suppose I have a .WAV with a song with some voice, some guitar, some bass and some percussions.
May I obtain back the single tracks from the wave, or is it an impossible task?

By the way, what is the algorithm to mix the tracks? Simply addition, followed by volume normalization?
Sorry if my questions are rather naive...
bilbo

hi bilbo!

what you are asking for is quite an advanced task. and "reversing" a wav is not easy, technically not 100% IMpossible, very much depending on your wav.

if you want to extract single "tracks" then you have some possibilities. i would recommend a good wav editor to you. like soundforge. or cool edit pro.

there are filters that ie filter by the panning of the signal amount. ie all that is plain in the middle can be filtered out. this is useful to find / extract voices. but depends on your wav.

you always have to cut out several ranges of the frequency spectrum afterwards ie normal equalizing (poor) or some configurable 24dB low|high|bandpass|bandblock filters.

a maybe better way to easily come to at least some result is to use cool edit pro's noise reduction filter.

you can mark an area of your wav, and generate a "noise profile" of it. that gives a frequency spectrum of that part. if you choose a part where there is only percussions then you will have the freq spectrum of "only percussions".

you can use this noise profile then and apply it to the whole wav. you can say either to cut out the profile or all else. meaning you get only the percussions, or all else.

please do not expect too good results. i sometimes was able to extract voices, but you need to remaster them. usually the frequency spectrum overlaps for the several "tracks" so you will never get a perfect result.
you can also work with compressors which wipe out all signals below a certain level etc. usually quite a komplex chain of different processing is required.

but try cool edit it might give you some quick success.

to mix, yes add them and normalize, the question is how (hopefully not in 16bit space) hehe

hope this helps as a quick start,

cheers, 0xf001

You may also want to consider searching for a midi of the song, which you can rip tracks out of directly.

Thanks, 0xf001 you told me some useful hints for a newbie like me...
I tried Noise Profile trick (FFT profiling) and FFT filtering with different windowing types (it seems that there are many algorithm for FFT filtering!)... the results are anyway not excellent, as you wrote...

I will also try to play with some direct sound manipulation and some simple program written by me, using maybe some sound manipulation package...

disavowed, your suggestion is like cheating! I was referring to some analogic sound obtained by mixing different instruments and voices and I was wondering if the mixing operation is really irreversible. Thanks anyway.

bilbo

Well, in theory it can be done but looking at the practical side shows that the amount of work is just more than insane. Mixing various audio sources into one result audio stream causes practically irrevocable overlappings and extinctions. In theory, you might FFT and get a complex system of equations which you just have to solve to get back (almost) the individual streams, but alas, quantum computers will be needed for this A poor but understandable comparison might be mixing paint... in theory you can separate the different paints again to get the individual colors back, but there is no practical way of separating them (don't come up with suggestions about making use of the different physical characteristics of the color molecules!)

hehe,

good comparison laola!

cheers, 0xf001

I friendly disagree. The human brain, for once, is able to abstract sounds from a complex, analog stream, like talking to someone in a rather noisy environment, you are able to clearly follow the other persons voice, and suppress the background. Also, a trained musician can focus on, and follow an individual instrument, even when the whole orchestra is playing simultaneously. The brain is not, as far as we know, a quantum computer.

The process of deconvoluting a complex stream of signal into the individual elements that generated it is routinely performed by computers. MRI and Echography imaging are good examples. Mass spectroscopy data is analyzed in this fashion.

naides,

i friendly disagre, too

the human brain can not filter in respect to get this part out. it can recognize and interpolate. this is a difference. in order to do the same thing with digital sound you have to
use another approach, we want to "extract" from a wav (also you can not hear a wav file, you can do some maths with it) hehehe

> and suppress the background.

that is what i mean. it does not suppress the background. it is always present.

regards, 0xf001

Well, okay, let me rephrase that "quantum computer" issue: You will need enormous computing power to get reasonable results. As you are mentioning mass spectroscopy and MRI... data of this kind is usually processed on computers that are very specialised and have much more computing power than a common PC Accepting your comment, let me cut a long story short: In theory it CAN be done but it doesn't really make sense because it is too much effort. Neither a trained musician nor the human brain will deliver the desired output bilbo asked for and I don't think he will buy a super computer to retrieve the original audio streams from the mixer result
Okay now? :beer:

I owe you a pint of Guinness

Howdy,

This is in deed one of those things that can be done but,
at what cost.

You could take a look at some of the stuff Cakewalk has.
Other then that, you will need to invest some serious money
to do what you desire. ($12,000-$15,000 US dollars)

Woodmann

You might find this article interesting: hxxp://www.soundonsound.com/sos/sep04/articles/qa0904-7.htm

In theory a waveform added to its inverse will cancel out. So if you happened to have the waveform of just the guitar in a song you could create the inverse of that and apply it to the original song thus removing the guitar. The problem is that the original waveform usually doesnt exist. Thats where the math and quantum computer comes in.

sort of off topic but an interesting article from a while back on removing digital watermarks from music:
hxxp://www.usenix.org/events/sec01/craver.pdf

Sounds cool
I'd suggest we all go over to the pub and first round is on me