Mid-side decoder removes stereo separation?

[neilren]

I have been running some tests in DCP-o-matic ahead of creating the DCP for a screening next week. The tests all start with a ProRes file with stereo audio, and use the mid-side decoder in DCP-o-matic to create a 5.1/HI/VI DCP in which all but the left, right and centre channels are silent.

Now, I only have stereo speakers at home and possibly that's the source of my problem and it's not really a problem at all, but...

When I bring the MXF files which DCO-o-matic has outputted either into Davinci Resolve or Audacity, something strange happens. If I play back all three channels at once (L, R, C) it sounds much like my original stereo ProRes. I can hear and see from the audio meters that there is stereo separation.

But if I mute the C channel and play just the L and R there is no stereo separation between them. I ran a test where I exported a ProRes with three very different audio clips running simultaneously, panned all the way left, all the way right and in the centre respectively; when I mute the centre channel of the resultant MXF I can hear both the left and right audio clips coming out of both speakers. If I solo the L channel I hear both the left and right audio clips coming out of the left speaker, and if I solo the R channel they both come out of the right speaker.

I am hoping that this is some quirk of the way Resolve, Audacity or MacOs (14.0) handles the playback of multi-channel audio through a stereo audio jack (my speakers are connected to my Mac Mini's headphone socket). But visually the left and right waveforms look the same, which also worries me.

Can anyone explain what is going on here?

[jimsky7]

I made my first DCP today also, and the audio sounds like mono and 'tinny'. I am playing back thru macbook pro speakers and various headsets. My original mp4 had 44.1k audio (I know, I know!) that DCP-o-matic said it was going to convert to 48k sample rate. I did not select any special decoder treatment. Cannot go to the theater to test - can't imagine it would work anyway.

I do have my film soundtrack as a 48k WAV, and if I could find a way to substitute that for the mp4's sound, I would gladly test it, but don't see a way to turn off the mp4's sound and use my WAV.

[jimsky7]

Just experimenting with settings: in DCP > Audio now I selected "Mid-side decoder" and "6 - 5.1" for channels, and now the playback on the Macbook pro within DCP-o-matic sounds perfect. Will send this to the projectionist for testing. I really like it to sound good on the Macbook before assuming the theater can somehow perform magic during playback.

[Carsten]

It will work better in the cinema, but there is a current flaw in DCP-o-matic that inverts Left vs. Right during the Mid/Side Process. This will give a very thin sound when played about close stereo monitor systems, and especially when you mute the center speaker.

I reported this to Carl, don't know when he has time to look at this. I would think it should be easy to solve once it's clear what happens during the M/S audio processing.

- Carsten

[jimsky7]

Good to know, and I will definitely verify the L/R swap you have indicated...it is critical as my audio is often spatially positioned to 'follow' actors on the screen. An actor far left will always be positioned more toward the L channel for example. I am listening to my DCP on a headset, so separation should be good and the tinny audio is quite clearly there - it's a giant effect, not just a slight effect. I burned another DCP with "6 - 5.1" rather than "8 -" yesterday and this sounds good to me this time.

[neilren]

Carsten, thanks. That explains it! Now I just have to decide whether to hope it does indeed sound okay in the cinema or play it safe and not use the mid-side decoder...

Has anyone out there noticed a problem with stereo separation in the cinema after using the mid-side decoder?

- Neil

[Carsten]

I had the idea to try this and fix the created Mid/Side MXF by simply loading it into Audacity, 're-invert' one channel, save, and create a new DCP with the fixed audio, then compare both. It's not overly complicated. A comparison in our cinema should be promising.


The way it is now, L/R will suffer 'some' cancelation in the cinema, which results in an emphasised center, but that is what most people would expect from a Mid/Side decoder anyway.

Currently I would need to find suitable sample footage for this. Need to browse my files. Maybe I should simply downmix a classic 5.1 trailers L/C/R to stereo for a try. I have plenty of these.

[carl]

I made a build adding a new "Centre channel creator" processor which does what I think you are expecting (left is left, right is right, centre is the average of left and right).

https://dcpomatic.com/build/make-centre/86c39b8

Let me know how you get on if you try it in a cinema.

[jimsky7]

I tried the Mac version just now. From listening on headphones within DCP-o-matic the L/R is correct and most definitely stereo with good tone and separation. However, the previous versiion of DCP-o-matic also did it correctly - at least for playback on the Mac inside the program. So I am not hearing any change. Not able to test in actual theater today. Will be doing that maybe during the week.

[Carsten]

I think the only issue with the current/old MidSide decoder is that one of either L/R is inverted against the other in the final result. This may be a residue of the inversion needed to extra the difference signal for the center isolation.

The mid separation itself is correct.

I think classic matrix mid separation is better than L/R averaging/summing at -6dB.


Carl - can you tell us how the current MidSide decoder works in the code? Do you use any existing libraries for it?


- Carsten