Topic: Hiss removal challenge

[it-goes-to-eleven]

I snagged a good show a couple of weeks ago from 3rd row center and have been struggling to reduce the hiss from a noisy analog stage. The recording is otherwise pretty good and I know of no others.

The hiss is broad spectrum and about 12 dB.   There is one un-amplified vocal where it is most noticable and which I also want to boost a bit.

My initial efforts with Audacity's Noise Removal function were encouraging (I used a sample of the hiss as the reference selection).. Until that vocal.  With a lot of noise reduction, the hiss would be gone but the vocal would be very metallic and hollow.   Settings that left the vocal intact removed only part of the hiss.  That result was *awful* because it left a digital artifact of the hiss algorithm (what was left behind followed a pattern).

I tried Sound Forge's paragraphic EQ but the results were so-so.

Are there any tools/techniques that would work without mucking up the vocal?

[heath]

send it to me and I'll try a few things...  no guarantees (are there ever guarantees in audio?    ) but I will try a few tricks....

[Karl]

Freelunch, how are you transfering the file to your comp?  Are you sure that you are not introducing noise at that stage?  What gear did you use to record with?  Also, keep in mind, when you are doing post editing, that if you add treble you will also be raising the noise level.

It is unlikely that you will ever be able to get rid of all the hiss.  You may be stuck with it.  Hopefully heath can do some good with it.

Heath, out of curiosity, what are you using to try and get rid of the hiss?

[heath]

i've plug in suites coming out the wazoo.  Lots of restoration/noise removal software.  We've tried them all, so I can try with each and see if any have a good result (some are great on one program, and terrible on others.  It all depends on source material and the noise in question.  There are also times when that I have works.  At that point, I send to a dude with a CEDAR system, but that's way too pricey for a Elvis Costello stealth....   

[F.O.Bean]

try soundforge/wavelabs de-noiser function, works very well IMO

[Brian Skalinder]

I was going to suggest playing around with some settings like (these from CEP 2.1): Snapshots in Profile, FFT Size, Noise Profile Plot, etc., all of which will have an impact on NR quality.  For my previous attempts, these three had the biggest impact, though others I haven't listed make a difference as well.  Unfortunately, in looking at Audacity's built-in NR utility, it's not very sophisticated.  So, two suggestions:

[1]  see if Sound Forge (since you already have access to it) has more sophisticated capabilities that will allow you to tweak the capture of your noise profile and its removal.  The CEP Help file is excellent, but dunno about the SF Help file - have you checked it?

[2]  try to find a more sophisticated NR plug-in for Audacity

[Karl]

i've plug in suites coming out the wazoo.  Lots of restoration/noise removal software.  We've tried them all, so I can try with each and see if any have a good result (some are great on one program, and terrible on others.  It all depends on source material and the noise in question.  There are also times when that I have works.  At that point, I send to a dude with a CEDAR system, but that's way too pricey for a Elvis Costello stealth....   

I have Samplitude as the main audio program and I run Waves plugins.  Have you had any luck with any of those tools?

[Brian Skalinder]

I was just reading in the 24bit FAQ that 16 bit material should be upsampled to 24 bit before any processing is performed.

FWIW, most editing s/w will upsample by default to 32-bit floating-point or 24-bit when performing DSP.

[F.O.Bean]

i have my soundforge and wavelab set at 32-bit floating for the DSP stuff

[Karl]

I was just reading in the 24bit FAQ that 16 bit material should be upsampled to 24 bit before any processing is performed.

FWIW, most editing s/w will upsample by default to 32-bit floating-point or 24-bit when performing DSP.

Because it's not clear to me, why is the 'floating-point' distinction always made?

[Brian Skalinder]

Because it's not clear to me, why is the 'floating-point' distinction always made?

Not sure if this will help at all:

http://service.steinberg.net/knowledge_pro.nsf/0/640f2835d6038517c1256baa004f6717?OpenDocument

[it-goes-to-eleven]

No word from Heath on this one, so I assume he gave up or was otherwise distracted by the impending Summer..

I spent a bunch of time working this over with Adobe Audition (just hiss work and a little level work).  AA is a decent tool and better than the others I have tried so far..  However, it is still unable to significantly redux the hiss without degrading the no-mic vocal.   I may ultimately seed two versions of that track - one with just bit of redux and one with full redux.  Probably worth it to hear the old man's un-processed voice for those who don't mind the hiss in that section.

I was just about to upload a sample of the no-mic stuff for anyone who wanted to check it out.  On a lark, I grabbed a sample of reference noise from the live vocal and did another redux attempt (the reference noise I had been using included the batbox but no mics). That seemed to work much better. That was in audacity so I'll probably go back and whack at it yet another time in audition..

This suggests that, if I am ever forced to use a noisy solution again, I should collect a noise sample with the mics somehow covered (the bathroom I setup in had a lot of ambient so those weren't suitable). Probably best to do that before I hit the venue.

Thanks again for the suggestions.

[heath]

my bad--i've been a bit all over the place recently.  I will take listen--I PROMISE!!! 

-T to me for being a slacker....   

[wbrisette]

I downloaded the flac files and the only problem I found was the no static file dropped way down in level. Is that suppose to be part of the problem? If not, can you make a better sample file for us to try and see what we can do?

Wayne

[wbrisette]

Thanks for taking a look, Wayne.  Those levels are part of the challenge!

He went completely off mic at that point.  Those levels need to be boosted by approx 8-12 dB.  That will raise the noise floor.  In the sample you grabbed, I left just a snippet of the amplified levels for reference.

There is some question of whether to raise the levels before or after the noise removal. There is also the question of whether you raise the level of the reference noise by the same amount (for cases where it is used during the NR process). From my attempts and thinking so far, the level should probably be increased before the NR.  Otherwise, you risk the NR algorithm not removing noise that is not now significant at say .5 dB but which would be after +10 dB.

The levels probably need a bit more than the 8-12 dB from what I could tell, I had them up in the 15 dB range, but as you noticed the noise floor was also raised significantly. I did have fairly good luck with raising the level using Waves L1 plug-in, then having Bias' SoundSoap as then plug-in. After fiddling around a while I was able to get a pretty good compromise. I'll send you a sample tomorrow AM.

Wayne