NR - on each track, or NR the mix?

I've been operating on a theory about NR, but have not tested it.  I'm working on a recording now that will be a good opportunity for testing NR techniques on a multitrack mix.  I thought I'd try to get some opinions bouncing around before I start posting samples/results from my test tracks.

I'm working on a live recording - I'll likely use just 3 tracks - an MS pair on the guitar and an overhead VOX mic.  This was a totally acoustic preformance in a very small venue and the mics were about 3' from the performer.

A/C noise in the room was not horrible, but I'm hoping to cut it back some.  What makes this performance a good test tracks is that she did a lot of talking, there are comments from the crowd, at times she was whispering, so NR artifacts have plenty of opportunity to be heard.

I have been operating on the theory that it's best to NR every track before mixing/effects.  The thought being that artifacts will not be the same in each track and blending them will help temper the effects of NR.  Also, I don't usually apply reverb to the master so NR on the mix would be post-'verb and I figured reverb would help smooth over NR artifacts.

I could also see it the other way, however.  By running NR 3x on the same mix I'm adding 3x the artifacts where a single pass would be cleaner.

What's your theory/practice?

I'll post tracks/mix before NR, after NR on each track, and NR on the mix.  It's early in the process.  I don't know what I'm going to find yet.  FWIW, I plan to use the exact same parameters for the NR process in all cases so process variation won't be a variable.

I have to say this is not something I've tried - I'd be more interested in getting the noise reduced before committing the performance to record.

That said, if face with a multitrack like this, I would use NR on individual tracks, for the very reasons you outline.

Come to think of it, I did do a job like this once - it was an 8 track 1/2 master that someone sent me to remix.  There was some desperate noise on a couple of the tracks and I used individual NR on those.

There are theoretical arguments in both directions (mainly to do with noise summing), but on the face of it, I think I'd go with the individual tracks too - for the same reasons. Although I haven't tried this, NR is reputed to work better if applied in a number of smaller slices than all in one go, and the approch of NR'ing individual tracks would appear to be heading in the same direction. Interesting to hear what the actual results of a test are, though.

Agreed with all.  Because the whisper-level details are so critical, I would do each separately as well.

I think the best argument for doing the tracks separately is that (with a bit of luck) it won't be necessary on all of them and any little artifacts which might appear would be nicely masked in the finished mix.

OK - here's the first round of samples - and maybe the last round depending on where folks want to take this.

I was pretty light on the NR - This is a very dynamic recording and it's going to stay that way.

The filenames are self explanatory.
(about 5MB each)
http://www.cheap-tracks.com/mp3/mixed_no_nr_dry.mp3
http://www.cheap-tracks.com/mp3/nr_on_mix_dry.mp3
http://www.cheap-tracks.com/mp3/nr_each_track_dry.mp3

Here's the NR settings I used (suggestions/critiques welcome)


Artifacts are (IMHO) light in all cases, but it's pretty clear to me that NR on each track is the right way to go.  FWIW, the noise sample was taken right after the "blue flash" - I notched the ringing strings out of the noise profile sample.

And for those who are interested, here's the entire earthquake story and enough of the song that you'll get an idea of Trina's style in something closer to a final mix (about 10MB)
http://www.cheap-tracks.com/mp3/trina_earthquake_story_beautiful_01.mp3

Out of curiousity, were there any obvious differences between a noise profile taken from the mix and one taken from an individual track?

Your results match with what I guessed might be the case.  I tend to find that NR works best the more specific (i.e. narrow band and similar amplitude) the noise you're working on.  My guess was that there would be minor differences in the noise  from track to track (down to mic type/placement, relative levels, etc) which would make the NR on the mix a bit more "general" and therefore more likely to create artefacts when removed.  As you might gather from my question above, I also wonder if results might have been slightly different using a different noise profile for each half of the experiment.

Bob

Why 5% on the spectral decay?  seems a long way from usual.

Also - 18db reduction is more than I try for on material like this.  However I am always confused about this factor's interaction with the reduction slider and curve.  Seems that one or more of them are redundant.

Will have to listen more closely at home for eval.

Why 5% on the spectral decay?  seems a long way from usual.

Higher numbers makes the noise level "pump" - and that bugs me.

Also - 18db reduction is more than I try for on material like this.  However I am always confused about this factor's interaction with the reduction slider and curve.  Seems that one or more of them are redundant.

 I agree, and I'm not sure I know the answer - I futz with settings and listen - futz some more and listen.  I press "GO" when I am either satisfied with the results or tired of messing around.

Will have to listen more closely at home for eval.

I was listening to the artifacts in the A/C noise when there isn't anything else going on.  I also looked for artifacts in very quiet audio.