Next gen stereo image tool teaser aka vocal removal

[Space Boy]

Nice work!

Hi,

Here's something I've been working on lately. It KILLS all tools similar to it.

Please try the original 'dusty' clip in 'Extra Boy'.

Better still, post it (.wav please) so that others can.

BTW, I just re-combined the two 'dusty' clips and 'Extra Boy' was in no way gasping for its last breath. In fact, I was pleasantly surprised how it managed to isolate even more of the vocal (leaving almost no background at all).

P.S. 'Extra Boy' runs real-time as a plug-in.

[Kingston]

Better still, you post the clip. (the mp3 artifacts in those clips are extremely minor, no need to worry about that.)


I probably never quite "got" extraboy as well as you (for fairly obvious reasons).

Besides, at least you can run extraboy realtime.

[Space Boy]

Better still, you post the clip. (the mp3 artifacts in those clips are extremely minor, no need to worry about that.)

Center signal (ie common signal) - supplied by Kingston.

Dusty Vocal Only (.wma) - Extra Boy

My mp3 encoder isn't working so I had to make a .wma file.

I'm not concerned about the mp3 artifacts from the original. I am concerned about combining the two clips you supplied. One is mono and the other stereo. I detect that a straightforward combination is missing some phase information and so my results from the combined files could be less than optimum.

Also, mp3 encoding can mess with the phase information. So I start from a less than favourable staring point.

Even so, not a bad job.

P.S. I used only the Isolate Vocals (HR) preset with no tweaking.

[fritzman]

Center signal (ie common signal) - supplied by Kingston.

Dusty Vocal Only (.wma) - Extra Boy

While this plugin removes the background better the remaining voice is full of artefacts. Kingstons stuff sounds way better.


Best wishes, FRitz

[autloc]

Keep in mind that Space Boy's stuff has been through two generations of coding (plus the two algorithms) - unless done at the highest bitrates, that's not kind to any audio.

[Kingston]

Space Boy that's a good demo of extra boy.

Granted, it is fairly FFT-watery (unless you used wavelets as well?).

By the way, My "dusty" demo is just a quick knock up using default settings. There are three parameters (width, depth, tension) for the extraction window, and if set right, they will leave no background noise. I can do uber-slicing if needed. The defaults are designed for even 2-to-3 division, not for slicing or isolating a particular sound.

Remember, what I'm demoing isn't "vocal removal" per se, but nearly transparent 2-to-3 channel unmixing. The "background noise" is what happens to fall under that particular wide(ish) extraction window, not an error. Next round of demos will feature my algo at its worst(distortion demo) and at it's sharpest(accuracy demo) to get things into perspective.

[autloc]

I've never heard of uberslicing before. Time to open a tab and look it up on citeseer?

Look forward to your future demoes, Kingston.

[Kingston]

I detect that a straightforward combination is missing some phase information and so my results from the combined files could be less than optimum.

Well you might be doing something wrong. Or are the mp3 artifacts really that bad? (could be, I didn't check that far)

A correct pseudo-code summing of the files,

Xsidesleft+center=left
Xsidesright+center=right

With the unpacked .wav that gives sample/bit/whatever-accurately and verifiably the original stereo file.

[Doug1]

your an idiot you come on here to tell us how good this is but we wont appreciate how good it is because we dont or wont understand whats involved.

Take a hike

[Kingston]

your an idiot you come on here to tell us how good this is but we wont appreciate how good it is because we dont or wont understand whats involved.

Take a hike

bleh. the discussion here has been intelligent until this little comment. *You* don't have to come here if you feel offended somehow. Heck, I frequent an electronics forum where the level of discussion goes way over me. It's not their fault I don't understand the discussion.

I'll answer to anything you or anyone else ask about the algo, but I suspect I've already told almost everything (not in laymans terms maybe).

This is the *DSP* forum you know.

[Watto]

wtf, Doug1?? Sure, Kingston can be obnoxious at times, and quite patronising, but you reminded me of this dog who used to live in the house next door to me - would jump up and go agro for no obvious reason...

[Kingston]

An extraction window border distortion demo. I found an example with an instrument that occupies the full stereo space. There's no way to slice the image without touching the piano:

Center signal (ie common signal)

Stereo sides signal

Notice how the rest of the signal stays artifact free apart from the piano.

(I checked the mp3 summing and there's little or no encoding artifacts in all of these demos so you should be able to sum to mp3-perfect original)

Next up, precision sample ripping.

[Doug1]

I doubt you understand quite what's going on.




Thats what was said by the self proclaimed genius

[Doug1]

wtf, Doug1?? Sure, Kingston can be obnoxious at times, and quite patronising

True knowledge is knowing what you dont know


guys like kingston seem to know everything and thats all theyl ever know.


enough said.

I havnt got time for obnoxious patronising people but they should be stepped on.

[Space Boy]

While this plugin removes the background better the remaining voice is full of artefacts. Kingstons stuff sounds way better.

Best wishes, FRitz

But doesn't isolate the vocals that well.

I am also at a disadvantage with respect to the source material. I have had to take Kingstons pre-processed data and combine them before using Extra Boy. Also, given the level of artifacts in Kingstons Lennon (imagine) demo, I am not conviced that a combination of the two parts yields the original signal.

To be fair, Kingston should post the original un-adulterated segment for others to make their own comparisons. However, copyright might prevent this.