De-reverberation - The final cut

[pluginnow]

Yes, I have done several improvements to the algorithm. I had VERY nice results with my "reference" track, witch is Bduffy's Vocal Plate.

Anyway, I'll do some tests, because I think I'm reaching the limits of the technology I'm using, and it's a lot. In any case, once I realize that it's done in terms of algorithm, I'll post that very strong plate, as clean as it's possible.

The file I'll post in some hours in absolutelly stereo. I'm using NO tricks like getting everything mono (or better than that..), or like using noise reduction methods. Stereo is absolulety preserved, just like the noise..

Several stages can be introduced in the algorithm later, it's just a matter of some more reseach. This is not an easy stuff, it includes billions of calculations.

The reseach I have done to reach this point, was very apreciable, and I'm on my limits.

And for those who asked, I'll post in public. I'm using 16384 points..

Thanks for listening

[pluginnow]

Greetings!

Instructions:

Listen FIRST the original file. It's a very strong reverb, very wet, difficult and extreme.

Then listen the de-reverbed file. It still have some residual sounds from the reverb, but thats modulation and also it seems to have a small pre-echo witch the algorithm isn't prepared to deal with yet (this pre-echo is also present in the original, I believe).

Here it goes:

http://rapidshare.de/files/35110544/de-reverb_kvr.zip

[northland]

The server is topped.

[bleebsen]

Artefact party

I'm afraid I'm not impressed.

[living sounds]

It's unuseable. The algorithm produces artifacts such as pre-ringing and noise-reduction-like resonances.

[pluginnow]

What would you expect with such a strong example?? Do you know anyone with better results with this file?

You post a resonable file, and then you can see what can be done with this technology, but no absurd please!

It would be nice to try a real example.. Give me a source to play with

[Rock]

I wonder how it would do with solely natural acoustic ambiance... I wonder if natural reverbs from a real room are more complex than a uniform process applied to an audio source. (I imagine that if a singer merely turned his or her head in a live room the mic would pick up different reflections)

[Branis]

It's unuseable. The algorithm produces artifacts such as pre-ringing and noise-reduction-like resonances.

Timestretching and denoising also produce artifacts when used at wide settings, but they're not unuseable. This effort should be encouraged, it's the first of it's kind in software (at least I don't know of any other), I saw once an expensive hardware de-reverberator.

[living sounds]

There was (is?) Prosoniq's pandora which cost 5.000 $ and had the same artifact problem (it used neural nets though). Still, I don't see much use in a plugin that switches reverb for artifacts. The great timestretching algorithms on the other hand are used because they produce very little of these artifacts. But, of course, this effort should be encouraged, I already said that I find it to work pretty good on the "It's Oh So Quiet" example.

[Krakatau]

- Could you eventually post example that are ( or are close to be ) usuable for a producing point of view ?

- In what cases is it the most effective ?

_________

By my side i would be interested of dereverberation of natural sountakes, even if it does not remove it completely, at the time the result is without audible artifacts

I assume that from one criteria, it would be easier because a natural soundtake isn't a combination of different simultaneous reverberation from different signal sources, as achieved productions of pop musics of all kind, usually are

_________

Hang on, many people here haven't much considerations for nothing else than something musically and immediately usuable !

_________

To those one, no offence, we may have here a young coder that deserves some encouragements ( That reminds me of Urs Heckmann presenting the early versions of MFM a few years ago now...)

[pluginnow]

- Could you eventually post example that are ( or are close to be ) usuable for a producing point of view ?

- In what cases is it the most effective ?

_________

By my side i would be interested of dereverberation of natural sountakes, even if it does not remove it completely, at the time the result is without audible artifacts

I assume that from one criteria, it would be easier because a natural soundtake isn't a combination of different simultaneous reverberation from different signal sources, as achieved productions of pop musics of all kind, usually are

_________

Hang on, many people here haven't much considerations for nothing else than something musically and immediately usuable !

_________

To those one, no offence, we may have here a young coder that deserves some encouragements ( That reminds me of Urs Heckmann presenting the early versions of MFM a few years ago now...)

Thanks all for the kind words. Anyway, I really really would like to test a "real" file, from the real world, in normal circustances.

Please just post such a file

Thanks again

[pluginnow]

I can't believe that no one have a file to post!!!

[kilroy]

Respectfully...I can't believe that the developer himself would not have suitable files at his disposal upon which to test his work.

[bduffy]

Yes. Pluginnow, why can't you post an example? You don't have Compact Discs where you live?

[camsr]

Okay I made a file to be tested of a natural reverb source. The format is 44100hz, 24bit, and please don't complain about the noise

http://www.yousendit.com/transfer.php?a ... 62d766e490