KVR Audio

Kingston · Post by **Kingston** » Fri Nov 04, 2005 11:10 am

Hi,

Here's something I've been working on lately. It KILLS all tools similar to it. It won't be released as a plugin any time soon though. The teaser is here just so you know what is possible with good stereo image slice extraction or removal (aka "vocal removal").

In this example I've extracted and removed center although I could've chosen any other image spot as well. (or use much wider or deeper scoop)

Combine both demos to the get the perfect original stereo file. It's basically near perfect 2 to 3 channel separation.

Center signal (ie common signal)

Stereo sides signal

This thing is a dream come true for sample rippers, or surround remastering. You could do discrete 2-channels to 3 reshuffling sample accurately while still preserving the original imaging. Or you could switch places of center and left part of the image for example. Or you could transparently remove or turn down (or up) that annoying tambourine on the "high left" of your signal.

This thing is pretty amazing for mix studies as well. remove the center and you can hear all kinds of otherwise invisible image related mixing trickery.

Kingston · Post by **Kingston** » Fri Nov 04, 2005 11:20 am

Here's another stereomix turned to a 3 channel mix.

Center signal (ie. common signal in center extraction)

Stereo sides signal

Again, combine them together to get the perfect original.

Kingston · Post by **Kingston** » Sat Nov 05, 2005 1:44 am

bump (check this out)

fritzman · Post by **fritzman** » Sat Nov 05, 2005 2:20 am

Hi Kingston,

this is stunning! Never heard it that good. What are your plans with it?

Best wishes, FRitz

Kingston · Post by **Kingston** » Sat Nov 05, 2005 2:29 am

fritzman wrote:Hi Kingston,

this is stunning! Never heard it that good. What are your plans with it?

Best wishes, FRitz

Sell it to TC or lexicon or similar company. No really, I'm going to try. If that doesn't work I'll give/sell it to some interested developer who can do a plug out of it.

Althought it's a really specialised offline type tool and might not work as a plugin (other than offline or reduced quality). Sure you could run it realtime, but it's a pig.

A3ntar · Post by **A3ntar** » Sat Nov 05, 2005 2:35 am

If that doesn't work I'll give/sell it to some interested developer who can do a plug out of it.

Hmm...I thought that you developped a tool yourself to do this? Or did you produce the samples with a wav editor using "new" techniques?

Kingston · Post by **Kingston** » Sat Nov 05, 2005 2:40 am

A3ntar wrote:
If that doesn't work I'll give/sell it to some interested developer who can do a plug out of it.
Hmm...I thought that you developped a tool yourself to do this? Or did you produce the samples with a wav editor using "new" techniques?

I develop in Matlab, and unfortunately I haven't got the C++ coding talent to translate this to a plugin. Even if I did, I'd rather see some estabilished company release this than some unknown in KVR like me.

I doubt you understand quite what's going on. Tell me how I could do this using any traditional techniques? It took me half a year of wavelet similarity algorithm development to get here. This was also my final year project graduating from university.

sqook · Post by **sqook** » Sat Nov 05, 2005 3:53 am

I really like the way that this algorithm sounds... very clean, only trace vocal elements left. However, I must ask.... why exactly did you start this thread here? You don't want any of us to convert your code into a plugin, nor do you have any interest in doing it yourself. What exactly are you looking for here.... feedback, compliments, or hype?

Kingston · Post by **Kingston** » Sat Nov 05, 2005 4:02 am

Just demonstrating what's possible. I know perfectly well what it does and doesn't do so no feedback needed.

Although there's a teeny weeny bit of algorithm distortion left there as well (that weird warble), the left over vocal you're hearing is actually the room or the plate reverb used in the mix. You see, what happens here is turning 2 channels to 3, not vocal removal. I just used the term because people associate these type of processes with vocal removal.

A3ntar · Post by **A3ntar** » Sat Nov 05, 2005 4:36 am

I doubt you understand quite what's going on.

Hey, I never pretended to understand the underlying aglorith behhind your field seperation to begin with.

Why do you seem offended though? Do you feel awkward because you are using Mathlab instead of C++? I hope not, because nobody is blaming you for that.

To me it is like the whole Synthedit controversy. Some people even commercially sell plugins made with Synthedit ( i am thinking of SpectraLive for instance ). While some people will NEVER touch anything that have been near Syhthedit, I for instance, don't really care if the effect/instrument is good ( like GreenMachineII ).

Why not make it in synthedit? would it be possible? what exactly are you looking for here?

Kingston · Post by **Kingston** » Sat Nov 05, 2005 4:51 am

well I was slightly offended by this:

Or did you produce the samples with a wav editor using "new" techniques?

but I also understand some people might not see the difference between this and M/S stereo.

And please let's not bring synthedit into this. While it might work for building interface it wouldn't even near touch the actual math behind this. It should be done in C++ if at all. It's easiest to do in matlab, which is why I used it in the first place. I know math and algorithms well, but I'm no programmer by any means.

I'm merely demonstrating the process here, not looking for help or any further development opportunities (not from KVR anyway). This process is something that would be best suited to dedicated mastering processors (like TC or lexicon) and they just might pack the DSP power for this.

Kingston · Post by **Kingston** » Sat Nov 05, 2005 5:04 am

More examples. (By the way if someone feels I've gone past the fair use of copyrighted material with these processing demos I'll take them down.)

A stereo signal split to three discrete channels. Again, combine the two to get the perfect original stereo signal.

Center signal (ie. common signal in center extraction)

Stereo sides signal

and one more.

Center signal (ie. common signal in center extraction)

Stereo sides signal

Enjoy.

A3ntar · Post by **A3ntar** » Sat Nov 05, 2005 5:28 am

[quote]well I was slightly offended by this:

Quote:
Or did you produce the samples with a wav editor using "new" techniques?
[quote]
I see. Well, my question was not sarcastic at all, seriously. I was seriously wondering if you created the sound with a wav editor that supports scripting or similar. The quotes in "new" was not to be sarcastic at all. Sorry you took it that way.

Btw, who could this be used in mastering? it seems to me that when you master something you can already seperate the tracks/fields in any way or shape. You don't need to decompose the song if you already have it in pieces.

Kingston · Post by **Kingston** » Sat Nov 05, 2005 5:53 am

A3ntar wrote:You don't need to decompose the song if you already have it in pieces.

Unless you do surround (re)mastering and like the choice or haven't access to the original multitrack stems. There hasn't been much need for this yet because it hasn't been possible in this high quality in the past. This is a hi-end tool. I doubt anyone in KVR would need this for anything else than precision sample ripping.

Also, I haven't demonstrated the other applications yet. You know k-stereo and orban image processing shapers? Well, they're good, but simply cannot touch this new process when it is fully utilised in a similar image shaping task in post production.

Compared to MS stereo tricks? well how about I rip a stereo track to two MS stems? yeah, it's possible with this. Wouldn't you say it gives a bit more power in highend remastering?

autloc · Post by **autloc** » Sat Nov 05, 2005 8:39 am

A google search for blind source separation and wavelets turns up a lot of hits and I've heard some pretty good results from time domain blind source separation methods, too. Your algorithm sounds good, no doubt - just wondering what you've done new? Any chance of a link to your thesis or a paper?

Next gen stereo image tool teaser aka vocal removal