Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[mwillems]

Hendrik, you asked me to evaluate the new volume leveling changes once the build was available.  Thoughts:

* It looks like it works perfectly for audio; tracks with positive volume leveling adjustments that would push them into clipping at 100% get the properly leveled when internal volume leaves enough headroom, and scaled down appropriately when internal volume is increased towards 100%.  It appears to be working perfectly and those tracks sound much closer in volume to other tracks.  I tried a dozen different tracks and all worked exactly as expected.
* For the downmix adjustment, it looks like volume leveling is now adding 7dB (to the extent there's headroom available to do so) to both 5.1 and 7.1 films when they're downmixed to stereo?  I tried six or seven films and got a +7ish dB delta from the volume leveling listed (sometimes it was 6.9?). That goes a long way towards making downmixed films have a similar volume to stereo audio on the same internal volume setting, but two notes:
1) The difference in perceived volume between stereo audio and downmixed 5.1 movies sounds closer to 10dB to me than 7dB, but I'll confess I'm only using my calibrated ear, so I may be full of beans. If the 7dB is mathematically derived, my ear may just be faulty.
2) Using the same delta for 5.1 and 7.1 appears to make 7.1 to stereo downmixes sound noticeably softer than music (and definitely softer than 5.1 downmixes).  That seems intuitive since you'd need to do more attenuation to downmix 7.1 to stereo without clipping?  

Takeaway:  overall a tremendous improvement, but for maximally level playback, I'd suggest differential boost for 5.1 and 7.1 downmixes and maybe a little more of it.  This is already significant progress on achieving perfectly level playback, though, and I really appreciate you taking it up!

[Hendrik]

The way downmixing usually works is simple. To avoid clipping when combining channels, the volume is reduced an appropriate amount for each channel that is added - depending on the peak level of that channel (from analysis) and the mixing factor (example: add two full channels at 100%, volume needs to be reduced by factor 2, or 6 dB)

When using internal volume and volume leveling, the volume leveling is now told how far the volume was reduced by the mixer, so that it can take this into account when leveling. With enough headroom, this volume reduction is reversed entirely.
So yes, these values are mathematically derived, and if all goes well and has enough headroom, the front left/right channels should still have the same volume as before the downmix, just with all the other channels mixed on top.

In practice that means that 5.1 and 7.1 should behave differently, as they have different amounts of channels that get mixed, and as such different amounts of volume reduction (which internal volume would restore).
I would hope that with enough headroom, they would sound similar, however. But, probably not the same.

Perceived equal volume when downmixing is tricky, however. For it to be perfect, our downmixing would have to be in perfect agreement with how the R128 volume level for multi-channel is computed (which effectively does a "downmix" of the individual channels volume levels to one global level), but its probably not, so the analysed value doesn't match the downmix anymore, and volume is slightly off.

[mwillems]

The way downmixing usually works is simple. To avoid clipping when combining channels, the volume is reduced an appropriate amount for each channel that is added - depending on the peak level of that channel (from analysis) and the mixing factor (example: add two full channels at 100%, volume needs to be reduced by factor 2, or 6 dB)

When using internal volume and volume leveling, the volume leveling is now told how far the volume was reduced by the mixer, so that it can take this into account when leveling. With enough headroom, this volume reduction is reversed entirely.
So yes, these values are mathematically derived, and if all goes well and has enough headroom, the front left/right channels should still have the same volume as before the downmix, just with all the other channels mixed on top.

In practice that means that 5.1 and 7.1 should behave differently, as they have different amounts of channels that get mixed, and as such different amounts of volume reduction (which internal volume would restore).
I would hope that with enough headroom, they would sound similar, however. But, probably not the same.

Perceived equal volume when downmixing is tricky, however. For it to be perfect, our downmixing would have to be in perfect agreement with how the R128 volume level for multi-channel is computed (which effectively does a "downmix" of the individual channels volume levels to one global level), but its probably not, so the analysed value doesn't match the downmix anymore, and volume is slightly off.

Got it.  I can tell you from empirical testing that both 5.1 and 7.1 downmixes both showed an almost exactly 7dB boost over their R128 volume leveling value when "unconstrained" by headroom.  I tested (in the end) about a dozen films, and all video I tested appeared to get almost the exact same amount of boost over their R128 value (within a tenth of a dB).  This was with internal volume set to -30dB so there was plenty of "room to grow." 

The 7.1 channel films sounded noticeably quieter than 5.1 channel films, and from what you say that may be inevitable?  But it sounds they should be getting differential boost, but I'm not seeing it.  Or is using the R128 value as the benchmark for "neutral" my problem?  Anything else I can do to help with testing?

[Hendrik]

The difference in 5.1 and 7.1 is usually not very big, because the extra back channels often are relatively quiet - and since we know their true peak values, we don't have to reduce the volume as much if we know that they'll never go over a certain peak.

In a sample I tested right now, 5.1 was using a 7.3dB reduction (with close to 0dB peaks in all channels), and 7.1 a 7.8dB reduction with < -5dB peaks in all channels, < -10dB in the rears.

If you can find a 7.1 track where all channels have a peak close to 0dB, then it should result in a higher reduction factor.

Regarding improving the perceived volume, well I'm not sure how to improve that further.
I might do some research, but if anyone has any knowledge that may help, do speak up.

Maybe JRSS needs a few tweaks here as well. It mixes fronts and surrounds with the same weighting, if we instead reduce the surrounds a bit, it may give a better sound image, as the front (which is usually the primary audio) is more dominant.

[mwillems]

Don't get me wrong, it's working fairly well right now (much better than it was), I'm just giving the feature a workout  .  I'll try and find some 7.1's with high true peaks and see what happens.  

Regarding improving the perceived volume, well I'm not sure how to improve that further.
I might do some research, but if anyone has any knowledge that may help, do speak up.

Maybe JRSS needs a few tweaks here as well. It mixes fronts and surrounds with the same weighting, if we instead reduce the surrounds a bit, it may give a better sound image, as the front (which is usually the primary audio) is more dominant.

If considering tweaks to the JRSS matrix, there's something about the weighting that's bugged me for years (in fact one of my first forum posts was about this issue).  The center channel content has always sounded too soft to me in the downmix when compared to the Left and Right channels (and sounds softer than dialog in stereo video that doesn't need to be downmixed).  Currently getting audio levels for downmixed video to where I can clearly understand all dialog results in the music (L+R content) being really quite loud.  Reducing the surrounds might help with that to some extent (less competing sound), but I would really like it if the dialog/music balance in multichannel video downmixing could sound more similar to dialog/music balance in stereo video.

Actually, come to think, that may be what accounts for the perceived difference in volume between what the algorithm produced (~7dB) and what sounds level to me (~10dB) as I was comparing volume of dialog in 2 channel video to dialog in 5.1 channel and 7.1 channel video.  I'll try using score/music volume to benchmark and see if I come to a different conclusion.

[mwillems]

In a sample I tested right now, 5.1 was using a 7.3dB reduction (with close to 0dB peaks in all channels), and 7.1 a 7.8dB reduction with < -5dB peaks in all channels, < -10dB in the rears.

If you can find a 7.1 track where all channels have a peak close to 0dB, then it should result in a higher reduction factor.

I found a 7.1 track with a 0dB peak, but I still only saw ~7 dB of boost over the R128 volume leveling amount.  I also did some additional testing, but I'm still not seeing significant deltas between 5.1 and 7.1.

That said, after spending some more time with it, the overall level of different content is much closer in general, so I'm very happy with how this worked out.