Topic: Fixing an off-center recording

[WiFiJeff]

A friend recently sent me a file of a nice piano recital he taped from the extreme right side of the house, with the sound quite definitely coming from the left side of the headphones.  The average volume for the left and right track is quite close, in fact the right channel measures marginally louder.  Clearly there must be timing issues here as well accounting for the perceived placement of the piano off to the left (which it was).  Is there any suggested delay that might correct this, has anyone successfully tried this?

[Gutbucket]

Likely just a millisecond or two, but whatever works.  Will need to listen and adjust for best effect.  Good technique to try 1st.

Alternately, I made some stereo recordings nicely listenable that were very annoyingly "everything of interest over to one side" by treating them as Mid/Side.  Assign the the channel with the stronger primary content as Mid and the other channel as Side.  Dial in ratio to taste.  Places the primary content channel dead center and splits the other channel out to either side.  Centers the off center content while preserving stereo room reverb.  The biggest potential problem is it can cause deep bass to skew left due to the right channel receiving the polarity inverted side channel.  However, that might not be a problem with solo piano, or an ensemble that doesn't have heavy deep bass content.

[boomfizzle]

I've used Gutbucket's suggestion for M/S in the past on recordings with problematic stereo images to great success, but before I go that far these days, I usually start with Izotope RX's Azimuth module. The Suggest button usually does a great job matching level (left side of module) and if I know I'm feeding it an off-center image, I'm more likely to let it suggest a delay (right side of module) as well.   If your fileset is broken into chunks, feed it your most consistent chunk and use those same suggested levels on all the fileset chunks. 
I never use Adaptive matching on the level side because the artifacts seemed weird to me on previous attempts.  I never use Adaptive delay unless the mic stand was really swaying during the show or something.   YMMV but it might be worth a render to hear for yourself. 

[voltronic]

Another thing you can try is to delete the first few samples from the left channel and shift it to line up the start with the right. Yes, this amounts to a delay, but now you're working at the sample level rather than the millisecond level. It may be just enough.

[WiFiJeff]

Thanks for the suggestions.  I ended up going with a 7ms delay to the left channel, which may have been overkill but I liked the centering this produced.  While fooling around I noticed that generally with centered piano recordings I was seeing a consistently louder right channel, which given that the sound is coming more from the open body off rightish and not the keyboard on the left is reasonable.  Raising the right channel by the observed 2.5 dB I measured from another centered recording did not help, however, the time delay seems to account for the perceived shift.  On speakers the whole thing is not a big deal, but on headphones (and I guess most people do a lot of listening this way, I certainly do) the difference is major.

[Gutbucket]

Good to hear it worked out.

Yeah, although a delay of millisecond or less should be technically sufficient (assuming typical mic spacing), I've similarly found in practice it often took more to really get things perceptually centered, frequently ending up with a similar delay to what you ended up using.  The only truly sufficient method seems to be adjusting and listening.

[Chilly Brioschi]

Well, historically, this:

[Gutbucket]

Following up from my spinning chair. 

Since this topic was about correcting a recording that had already been made this is somewhat OT, but the question of how to best accommodate an off-center recording position before pressing the record button is related and has been discussed at TS in the past. 

The simplest method is to setup as normal, then close your eyes and try to forget where everything actually is while turning to face the apparent acoustic center from which the music seems to emanate.  For PA amplified concerts, that's going to be somewhere near the closer PA speaker.  Rotate the mic stand to point the mic array directly at that "acoustic" center point, even though you may be strongly tempted to point the mic array toward center stage or toward the opposite side PA in an attempt to compensate.  Doing that will only aggravate the imbalance.

^ That's really all I wanted to follow up with. 

But for anyone interested.. There ARE ways of pointing the mics more toward the opposite side, however it requires an asymmetrical microphone configuration that is not obvious or easy to determine, where the baseline between the two mics is angled differently than microphones.  That makes it a far more complicated approach which is not likely be useful for tapers.  Basically it uses time-of-arrival / level difference "trading" and the directional pattern of the mics to skew the image and energy distribution to one side or the other.  The excerpt below is from one of Michael William's papers hints at the underlying relationship.  We can discuss how to practically apply it to taping if anyone is interested.

[ol' dirty taper]

I pulled a tape recently where I had a configuration very similar to Fig 11b, where I was off center about 10' and did what I considered at the time a modified PAS with mic pointed just outside the stacks, just skewed. Was very happy with the results.

[goodcooker]

I set up a lot like fig11b quite often. Depending on how far back I am and how directional the mics are. Hypers = just like that, subcards = more typical arrangement since they are much more forgiving.

[Gutbucket]

Somewhat confusing that 11b and 11c are identical, only rotated differently.  Both time and level offsets are happening in both cases.

I haven't done this much, but have used arrangements like 11c a few times to push the mic pointed toward the more distant-side farther forward, in an attempt to pull the image toward that direction by leveraging the offset in time-of-arrival.  Doing it that way is easier for me to visualize and figure it out in the moment.  Its essentially doing the same thing as the simpler method of rotating the entire stand to face toward the apparent acoustic center, but adds the additional change in individual mic angles.

Pushing the distant-side mic farther forward works regardless of pattern since its a time-of-arrival based thing rather than a pattern/level based thing.  Making an additional level correction afterward is easy to play around with, but trying to arrange for such a level difference via microphone pattern and angle is tougher to wrangle, because any angle between the two mics tends to works against getting increased level from the opposite, more-distant side along with decreased level from the closer/louder side. Because of that, I've tended to arrange the mics more parallel to each other than I would from a more centered recording position - more A-B like, with both pointed toward the PA on the opposite side, using additional space between them to compensate for the reduction in angle and to increase the useful time-of-arrival shift.  That uses pickup pattern directivity to increase sensitivity to the far side PA and reduce sensitivity to the closer/louder nearside PA across both channels, without changing the overall energy balance like making a level adjustment afterward will do, and assists the time-of-arrival offset in shifting the image toward the far side, rather than working against it.

[Chilly Brioschi]

Fig. 11 seems to have been created for recording from the "Jerry side" of the stage.
Time and level can be tweaked in post as well
https://origin.izotope.com/en/learn/5-ways-to-adjust-phase-after-recording.html

[WiFiJeff]

I'm rebooting this thread because I've come up with another, more complex, centering issue, with no idea where to start. 

I recently revisited a recording from a performance from almost 20 years ago, of an opera with piano accompaniment (the composer's original piano score).  The performance was quite good, the voices well-trained and perfect for the small theater in which it was held, and it cleaned up nicely now in Rx10.  However, the piano was set not in a central position but rather far off to the left, and the quasi-binaural recording reflects this very accurately.  The voices alone are correctly spread across the sound stage, and although the piano is recorded on both stereo channels it is clearly way off to the left (this actually helps keep the voices nicely distinct, but it is distracting in its own way).  Any of the actions I have used in the past to move a single piano towards the center would here squeeze the voices way over to the right.  The mono mix is listenable but, well, flat.  It would be great if I could isolate a piano stem from the vocals and mix a mono piano with the stereo voices, but I can't get the Rx10 Music Rebalance to separate piano from voices (overlapping frequency ranges, is it doable??)  Does anyone know any way to do this, settings in Rx or another program to separate voices from piano, or any other method to accomplish a fix?

Jeff

[Gutbucket]

Was thinking Music Rebalance would be your go-to for this too, except for the problem of separating the sources.  Hopefully others may have suggestions which help with separation.

Lacking that, you might try a few different variants on the mid/side technique mentioned earlier:
1) To preserve the existing center image, add some polarity inverted Left channel to the Right.. and tweak stereo balance as needed to compensate. Since piano is primarily left in the original recording, doing this will spread it out somewhat diffusely to both sides.  The existing center image should remain more or less the same.
2) To use the existing Left or Right channel as center image, assign that as Mid, and the other channel as side.  Adjust ratio to taste. Try it both ways - Left as center vs Right as center.

[voltronic]

Here's a possible fix I just thought of, but it would be painstaking:

Use Spectral Repair to remove as much of the vocals as possible. Use the paintbrush selection tool and the Attenuate or Replace sub-modules. I recently did this to save a recording where the first violinist in a chamber orchestra had a persistent cough throughout an entire Requiem mass.

Export your result which should now be a (mostly) piano-only track. Now create two stereo tracks in your DAW: The original mix and the isolated piano. Pan the piano track in the direction you want it to go, starting with its level low in the mix and increasing until the piano image is pulled over towards center.

You might even try using your exported piano track to phase-cancel the piano in the original and get isolated vocal and piano to do with as you wish, but I suspect the vocal track will be mangled by doing this.