Subject:General usage questions

Posted by: Paul Masters

Date:8/6/2012 2:50:38 PM

Hello:

I have edited audio with Cubase (a number of years ago), Sound Forge a bit, but mostly with Vegas (and video). What I do is very simple. It is for my own use.

I am sure that the people here are much more experienced in audio than I am. I hope someone can give me pointers as to what I am doing wrong or need to do.

I saw SpectraLayers at NAB. I realize that the sample chosen was because it was easy to 'see' the problems and select them. However, I hoped that the product would help with some problem audio I have.

The audio is commentary mostly recorded outside with a Cannon XHA1. The clips are 44,100 Hz 16 bit. I know that the demo will resample them.
Perhaps that is part of my problem.

In the past I tried to remove the noise with Waves Restoration X Noise and a little EQ. However, the voice has a 'hollow' sound, like an old telephone effect.

Using the default settings in SL, I selected an area that was only noise and then 'painted' it over an area of the clip. (X Noise is easier to use. First there is a learning / setup step and then the profile is applied to the entire clip - no 'painting' required.)

That removed a lot of the background, but the voice was 'hollow' and there were a lot of 'singing robots'.

I then tried to isolate the voice. While I can hear the voice I can't always 'see' it in the display. One clip with a woman's voice has more than the default 10 harmonics - changed that to 20.

That resulted in some of the voice being there sometime and not others. I know that was because I couldn't see all of it to select it.

I am sure I am doing something wrong. Or perhaps it is the down sampling that is making these clips difficult if not impossible to process.

This appears to be a very useful product. And, I realize that not everything can be 'fixed in post' (G).

Thanks for any ideas.

Paul Masters

Subject:RE: General usage questions

Reply by: MarkWWW

Date:8/6/2012 3:52:17 PM

Perhaps if you can post a short example of the audio you are trying to denoise some of the people here can have a go at it with their preferred denoising tools/techniques. If any are an improvement on what you have been able to achieve then you can try using their techniques instead of yours.

It's a bit soon for too many people to have become familiar with SpectraLayers but there are several people hereabouts who are able to work wonders with the "conventional" noise reduction tools, and particularly with iZotope RX.

Mark

Subject:RE: General usage questions

Reply by: Paul Masters

Date:8/10/2012 6:02:36 PM

OK. I have a sample. How do I upload / post it? I don't have a WEB site or any
other URL to reference.

PS. Thanks to all who mentioned iZotope RX11 here. I searched for other products but did not find any. I am 'search challenged' (G).

PPS. The update to SL will read a 96/24 bit stereo file apparently with out resampling. At least it didn't show the 'resample' message.

Thanks.

Paul Masters

Subject:RE: General usage questions

Reply by: MarkWWW

Date:8/11/2012 12:03:42 PM

> How do I upload / post it? I don't have a WEB site or any
other URL to reference

If you don't have any webspace, etc, of your own you can use a file hosting service.

In the past I have usually used to free version of YouSendIt which I think is still available. Many people these days seem to prefer Dropbox. Google have their own Google Drive offering, and there are probably dozens of others - seems like that anyway.

Pick whichever file hosting service takes your fancy (the free accounts should be sufficient - no need to sign up to anything that costs money), and upload your file to the storage that they provide. Then let us know the URL which the hosting service provides for that file so we can download it and see what improvements we can make to it. (You may need to make the file available to others by ticking a "share" box or something similar, or the file may only be accessible by you.)

Mark

Subject:RE: General usage questions

Reply by: Paul Masters

Date:9/3/2012 3:48:46 PM

Hello:

Finally got the time to make 2 samples. One is of the noisiest and hardest to 'cleanup' clip I have. The other is a womans voice. Both are very hard to 'see' in SL.

I may get a demo of RXII when I get a chance and try it.

Thanks for any pointers using a 'reasonably' priced product.

Here is the link:
https://skydrive.live.com/redir?resid=FC09A1590A5424F8!135&authkey=!AAHDfUfxOzBhuks

Thanks for any comments. Even if it is "You're crazy! It can't be done".

Paul Masters

Subject:RE: General usage questions

Reply by: ChristoC

Date:9/4/2012 2:08:55 AM

The first file is very sad - signal to noise ratio is almost equal, so there's not much that can be done; SpectraLayers helps with some of the squeaks, but not the noise; iZotopeRX2 helps with some of the noise, and deals with squeaks adequately, but eventually kills the intelligibility if the voices if pressed too hard to remove noise. Second file is almost as bad. With both you will get easier & better results with iZotopeRX2, but don't expect much of an improvement. I thought Cedar DeHiss (a plugin for some DAWs) worked with the second file OK, but only removed the high end of the general broadband noise.
For these types of files you probably have to resort to some very big (.... read expensive!) bazookas e.g. = Cedar DNS-series hardware processors (http://www.cedaraudio.com/) but even those may struggle....

Subject:RE: General usage questions

Reply by: Paul Masters

Date:9/4/2012 11:24:14 AM

Hello ChristoC:

Thanks very much for taking the time to look at the samples.
I was afraid of that. They are very noisy.
I will try RX2 demo when I get a chance.
I know I can't get them completely clean, but was hoping for better than I have.
As they say, somethings just can't be fixed in post.

Thanks for your helpful comments and pointers.

Paul Masters

Subject:RE: General usage questions

Reply by: divide

Date:9/4/2012 4:53:40 PM

Hi Paul,

I had a look at your samples: first change the window resolution from default 2048 to 8196/16384, you'll have a much more accurate view of your spectrum in this example.
Next on such a noisy sample you should rather try to extract harmonics than to remove noise. I suggest you extract the rough harmonics from the voice, as much as possible, then resculpt using the eraser tool to get back the original formant shape of the voice; this way you'll be able to recover the voice without the noise. But it requires work, it's not an automatic process...

Subject:RE: General usage questions

Reply by: divide

Date:9/5/2012 3:24:09 AM

I gave it a try with SpectraLayers on the first seconds of sample2, have a look here:
https://rapidshare.com/files/496650889/Sample.zip

I worked with a window size of 8196, did a rough extraction of the lower harmonics using the extract harmonics tool, copied the upper harmonics and noises with a simple rectangular selection, then resculpted the lower harmonics using the eraser tool to remove the unwanted false harmonics (you can do a quick compare by disabling your voice layer, and quickly switch between your voice layer and the original mix using the shift+tab shortcut).
This is a long process but it will give you by far the best results over automatic noise filters you can find in other products.

Subject:RE: General usage questions

Reply by: Paul Masters

Date:9/9/2012 10:37:24 AM

Hello:

Thanks for your time and great work.
The sample you sent sounds amazing!
I will give your suggestions a try when I can.
However, I think your ability and patience is much greater than mine.

Paul Masters

Subject:RE: General usage questions

Reply by: Paul Masters

Date:9/15/2012 10:36:29 AM

Hello divide:

Thanks for the SL project. It was very helpful to see what you had done.

I am not sure about some of what you describe.

You mentions changing the 2048 to 8196/16384. I don't have those numbers in the pull down box at the top of the screen - perhaps you mean something else.
I did zoom in both vertically and horizontally and adjusted the 'brightness' of the display.

I used the harmonics tool, but many times it did not select the frequencies above the one I selected. If I moved the indicator below the lowest visible frequency - nearly or at the bottom of the track, then it would pick up more along with a lot of 'background'. Perhaps that is what you meant by using the eraser tool to remove the unwanted false harmonics.

I was able to get a few words nearly as clear as your sample. But some were still garbled so I gather that not all the harmonics had been selected.
While I haven't tried it yet, I thought about using the frequency tool to select each harmonic. I know that will take longer, but it may work better than the harmonic select tool.

I am running the trial. It is now on a Vista 32 bit PC as the trial on the W7 64 bit PC has expired. Perhaps it being the trial or the slower PC is causing the difference between what I see and what you describe.

FWIW... As a comparison, I got a trial of RX2. I found it much more intuitive than SL and much easier to work with 'out of the box'. I was able to clean up sample 1 that I provided with the voice sounding much better than the one I did with X Noise.
From your work with sample 2, I think SL can do more once / if I can get the hang of using it.

Any comments or suggestions will be appreciated.

Paul Masters