it is possiable to make a filter that will filter out only the human voice without filtering out any other audio ranges. id test it myself but dont realy have the equipment on me and at home i dont have a o-scope.

 

Probably - that may be the way karioke CD's get made. That is just removing the voice, not the vocal frequency range. If you take out the frequency range of the voice, there won't be much music left.

 

First of all, karaoke is typically done by removing the tracks from the master recordings and remixing.

You can remove the voice, but anything else in that frequency range will also be removed. I'm sure you can do better if you use some sort of digital signal processing, but I doubt it would work as well as you'd like, especially with the amount of work it would require.

 

If the voice is in mono like most are but the rest is in stereo then invert one channel and add them. Anything in mono (including most bass sounds) will cancel. The result is something like an echo chamber but some people call it "ambiance".

 

If the voice is in mono like most are but the rest is in stereo then invert one channel and add them. Anything in mono (including most bass sounds) will cancel. The result is something like an echo chamber but some people call it "ambiance".

never much understood the differences between mono and stereo to be honest with you, nor have i done much research on the matter. to be honest im bogged down with so many ideas and no where to start on any of them that my heads spinning.

 

If I were to approach this problem I would look at using an adaptive filter developed on a DSP. The more money you are willing to spend the better chip you can get and more effective your filter could (I say could because you still have to properly implement the filter) be.

Dave

 

There are a few machines that claim to remove vocals from audio tracks, but they're very expensive and they don't work on all songs.

Here's a machine that makes those claims. I'm willing to bet it's not cheap.
www.ltsound.com

Like Dave said. DSP is the only way of even thinking about solving this kind of problem.

Let's say you did an FFT of a block of audio data. You would have an array of amplitudes at certain frequencies. (We'll disregard the phase information.) You could scale (reduce) the amplitudes of the signals between say 300 to 3khz, (perhaps by a factor of say 2), and subtract the result from the original Frequency domain signal. Then, convert it back to time domain. You wouldn't want to remove the signals completely; this would leave a "hole" where the vocals were.

Then, you'd have to figure out what harmonics actually belonged there, (ie. harmonics from instruments that were mixed with the vocals.)
and put them back at the proper amplitude, or close to it.

It sounds simple, but it's an incredibly difficult task, even with human intervention. As far as "seeing into the future" goes, if you have a recording, you have apriori knowledge of what the next sound block contains; just read ahead by however many samples you want.

I won't say "it's impossible", because I've seen some amazing things done by people who were told that what they were trying to do was impossible.

(Ray Damadian comes to mind...)

 

The demos from IT Sound have extreme stereo separation for the background music and the volcals are in mono. So it is easy to invert one channel and add them to cancel the mono sounds. The songs don't have much bass anyway so you don't notice its cancellation.