X-curve compensation re-EQ

[SME]

This thread is more about full-range content than bass, but it is content related, so I think it works best here.  In the future, I may post this somewhere on AVSForum, but for now I want to keep it to a limited audience.

As I've mentioned in the main LF Content thread, the X-curve calibration standard in cinema causes two major problems:

• Tonal balance that deviates substantially from neutral and from what is typical used (informally) for music production and what sounds good on a home system that is optimized for music.
• Inconsistent calibration between different dub-stages and cinemas.

As I also noted, many UHDBR/BD/DVD releases these days have high quality home remixes that fix most of these tonal balance problems.  This is true for most recent Disney releases these days (including, e.g. the new "Star Wars" and much recent Pixar and Marvel stuff).  However, much legacy content as well as lesser quality home-remixes do not feature any re-EQ and retain the inverse-X-curve signature.

The effect of X-curve calibration is to attenuate both high frequencies, via the -3 dB/octave slope in power response, and the low frequencies, which arises from forcing a flat power-averaged response even though virtually all speakers have a significant drop in directivity for low frequencies and what absorption is present in typical dub-stage / cinema rooms is also less effective at low frequencies.  As a consequence of the altered tonal-balance, most mixes are likely altered to sound good in the dub-stage during the re-recording mix process in which highs and lows are boosted to compensate.

The resulting mixes, in addition to translating unreliably between theaters, sound less than optimal when played back on a home system.  The auditory symptoms are mixed.  I find it easiest to hear the problems in the dialog.  Sometimes only one of the excess highs or the excess lows is audible in the unaltered track because the boost dominates.  For example, some cinema mixes, the dialog comes across very bright.  In others, it comes across very boomy.  Sometimes, the dialog seems relatively balance, in terms of high vs. low, but with the mid-range being relatively depressed, intelligibility often still suffers.  Dialog is both much easier to understand and much more enjoyable to listen to when it's presented neutrally.

Unfortunately, the required correction varies between track for both of the above reasons.  Mixers don't necessarily attempt to defeat the X-curve alterations in any systematic way.  Instead, they "turn various knobs" and listen until they are satisfied with the result.  So the ideal filters to reverse their changes may vary between mixes.  And because the X-curve calibration method isn't even consistent between dub-stages, EQ-adjustments that give good sound in one dub-stage may not work well in another.  In fact, there's evidence that X-curve calibration doesn't even achieve consistency between the left and right vs. center screen channels vs. surround channels in the same dub-stage.  The situation is a big stinking mess for sure.

Nevertheless, even if the adverse effects of the X-curve standard on the mix cannot be perfectly reversed, it's possible with some rudimentary EQ to improve the sound quality of cinema mixes considerably.  Now that I've finally achieved a stable, reliable audio reference in my own sound system, I've been giving attention to this problem.  In this thread, I hope to document some of the candidate corrections that I've applied to improve the sound quality of various movies.

I would encourage anyone with the required capabilities to give these a try and share feedback.  To implement these requires the ability to apply various biquad EQ filters such as high and low shelves and Peaking EQs, ideally to the streams *before* bass-management.  Though for my first pass, I'm applying the filters identically to all channels, so it should work fine to apply them after bass-management as well.  One issue I imagine most people will have is that they have a limited number of free filter slots.  The more filters used, the better quality correction that's possible.  I will try to limit the filters to what's actually needed.

Edit: I posted a candidate correction for "Wonder Woman".  Sweet!

Edit2: TL;DR: I don't recommend using any "corrections" that I posted here.  This was mostly a failed experiment because most tracks don't exhibit a predictable inverse X-curve signature. Instead it would seem that typical EQ adjustments are typically smaller and less stereotypical.  Furthermore, EQ problems seem to be inconsistent within different parts of the same track (not just by channel), which means improvements to some aspects of the track are likely to come at the cost of degrading other aspects.  I also discovered that it was extremely difficult to get the shape right in the upper mid range, with nasty resonances resulting from errors.  Even now with much more precise response optimization for my system, I'm skeptical that any improvement I get with EQ will reliably translate to other systems unless I somehow manage to neutralize the original errors very precisely.  To do so would require more skill and better tools than I have.  This isn't just *my* problem but seems to be an issue with many of the home releases that do sound they like were re-EQed (e.g. many recent Disney).  I very often hear upper-mid resonance issues (albeit mild) on these titles, and I rarely hear such issues with mastered music.

[maxmercy]

Pinned.

JSS

[SME]

X-curve effect on cinema response:

Below is an illustration of the effect of X-curve calibration in a "typical" cinema.  The X-curve illustrates the X-curve target itself.  The black curve illustrates the power response that one might measure for a typical anechoic flat speaker.  The blue curve illustrates the EQ that might be applied by the calibrator to force the response to conform to the X-curve target.  Note that every cinema is different, so in reality, the black and blue curves will differ quite a bit due to differences in speaker directivity vs. frequency, effects of nearby boundaries, distance to the reference position, and room acoustics:

My work here is based on a key theory about audio reproduction and perception:  Our tonal balance perception correlates most closely with the anechoic response of the speaker, despite the other effects that shape measured in-room responses.

This theory ignores the effects of a movie screen which typically absorbs a lot of high frequency sound and effectively changes the sound of a flat speaker to make it rolled-off.  An ideal cinema calibration method would only correct the screen roll-off, but the technology to measure the effect of the screen in-room independently from the rest of the response was not available at the time that the X-curve standard was conceived.  Only power response measurements were available in the 1970s, which is, in large part, why we got a flawed standard.  The black curve above is an example of what an in-room power response might look like for a flat speaker with the screen effect precisely corrected for.  Likewise, the blue curve is not meant to represent the actual EQ applied to the cinema speakers, which in reality would have less high frequency attenuation or even some boost to overcome the screen.

Because I ignore screen effects, the blue curve is a rough and approximate illustration of how typical audio content (i.e. excluding cinema mixed content) will sound when played in a dub stage.  It is reasonable to expect the re-recording mixers to adjust the tonal balance by approximately the inverse of the blue curve.  And in order to reverse these adjustments for playback at home, we want to apply EQ that's similar to the shape of the blue curve.

I think it's worthwhile to note a few of the different frequency regions:

Above 10 kHz or so, most of the roll-off shown in the black curve arises from the effect of air absorption over distance.  It is meant as merely a rough illustration.  In reality, the air absorption roll-off is smoother and drops off faster closer to 20 kHz.  This example is a reasonable approximation for what happens at about 10 meter distance.  I have experimented with high frequency roll-offs that accurately simulate the distance effect and find that they have only very subtle effects on the overall sound until the attenuation gets very high (like 10 dB or more).  As such, YMMV on this part of the correction.  Each listener want less or more, according to preference, or you might not hear a difference one way or the other.

Between 500 Hz and 10 kHz, the black curve exhibits a -1 dB/octave slope.  A flat speaker exhibits this shape due to a combination of factors.  (1) Speaker directivity often decreases slightly with descending frequency over this range, meaning that speaker power response may be a bit higher at the low end of this range.  (2) Room surfaces often absorb a bit more up high, resulting in a slight decrease in reverb decay time and less build-up of energy at higher frequencies.  (3) Air absorption effects high frequencies (much) more, resulting in decrease in reverb decay time.  All three of these effects are usually quite subtle over this range; hence, the slope is pretty mild.  Note that in the real world, the slope is often flatter in some parts and more sloped in others and varies between speaker and room. 

Between 100 and 500 Hz, the in-room response of a flat speaker rises more rapidly from two factors: (1) Speaker directivity drops off rapidly with descending frequency lead to a big increase in sound power in the low frequencies.  (2) Room surface absorption becomes much less effective leading to a big increase in decay time and reverb build-up in the low frequencies.  Note that the inverse of the blue curve also drops quite a bit, leading mixers to frequently boost the low frequencies quite a lot on the soundtrack to compensate.  Unfortunately, this demands a lot of headroom in the mix.

Not depicted is the range below 100 Hz.  Below 100 Hz, room decay time typically continues to increase, dramatically in some rooms.  Peaky room modes may develop and dominate response.  The X-curve target itself slopes down as well, which leads to substantial and very inconsistent attenuation of bass in the dub-stage playback system.  As noted above, compensation for this attenuation in the mix itself is common but is ultimately limited by the headroom.  As a consequence, cinema mixes often have weak sounding bass when played in X-curve calibrated cinemas, even with the help of the LFE channel.  For this reason, I rarely feel the need to attenuate the bass beyond what is accomplished by the shelf in the 100-500 Hz range.

What a typical correction looks like:

My generic correction involves essentially 3 or 4 different components, from top down:

Ultra High Frequency / Distance Roll-off:

I already use a distance-emulating ultra high frequency roll-off shelf for music playback, which these days is set to roll-off about -3 dB @ 20 kHz from flat.  For cinema mixes, I have it roll-off around -7 dB @ 20 kHz from flat.

High Frequency Slope:

I attempt to mimic the high frequency slope of the X-curve target over the 2kHz-10kHz range using two shelves.  However, I find -3 dB/octave to be too much in most cases.  I find -2 dB/octave to be a good starting point.  I recently introduced a EQ peak filter to sharpen the knee around 2 kHz that otherwise dips below target when using shelves.  I haven't had time to experiment with whether the sharper knee sounds better or not.  And of course, what sounds best could easily depend on the mix because we're not exactly trying to invert the the X-curve but rather invert the compensation the mixers did because of the X-curve.

Low Frequency Shelf:

I shelve down the low frequencies with a center around 200-300 Hz, usually between 3 and 5 dB.  I start with a Q of 0.625.  The most important thing to listen to when optimizing this shelf is the dialog, and I like to listen to speakers with voices that cover a few different pitches.  Too much low frequency causes muddiness and/or boominess of the voice.  Too little low frequency doesn't usually impact intelligibility, but it doesn't sound as natural.  Furthermore, I find that there's a kind of anchor point, in and around 180-200 Hz, where the bass and the mids and highs meet.  Too much in that region muddies things out.  Too little in that region causes the bass (even a lot of the low stuff) to lose much of its impact (tactile or otherwise).

Mid Frequency Shelf:

The mid frequencies may also benefit from a slight shelf, between 0 and 2 dB somewhere between 500-2000 Hz, to better balance the mids and upper mids.  The shelf should increase in the direction of the upper mids (see the blue curve above).

Wrapping it all up

In the end, these corrections are guess-work, and the only sure way to know what works best is to experiment and listen.  I call these candidate corrections because better corrections may be possible with more time and effort spent.  In the future, I hope to have the capability to make EQ adjustments live, so I can hear the results immediately and make interactive adjustments.  That should allow me to achieve much better results much more quickly than the process I'm using now.  Nevertheless, it's remarkable how little effort I need to spend even now to close 75% or more of the gap in sound quality between the uncorrected mix and the ideal re-master (whatever that is).  Surely a pro could re-EQ and make a huge improvement to sound quality in no more time than it takes to run through the film.  I'm fairly certain some studios are already doing this.  But others are not, either because they are not doing home re-mixes at all or because they are simply not aware how much these remixes can benefit from re-EQ.

As a complicating matter, tonal balance is not consistent between playback systems.  I have worked very hard to discover a reliable reference tonal balance response and to develop a systematic means of calibrating my system to it.  I'm not going to try to describe that process in this thread.  I've written about it elsewhere.  Though I've refined my approach further since my last comments on the subject.  It's getting a bit difficult to fully describe my approach on a forum as it's getting quite technical.  In any case, just because the movies sound better with these corrections on *my* system doesn't mean it will on everyone's system.  However, part of the idea of having a good neutral reference is that a given re-EQ should still sound better on most peoples' systems.  Any feedback anyone can offer is welcome.

[SME]

Wonder Woman (2017, BD / Atmos)

UHF / distance (relative to flat response; see note*):

High Shelf: f0 = 13900.0; Q = 0.4675; gain = -3.501

HF slope:

High Shelf: f0 = 3140; Q = 0.707; gain = -0.5
High Shelf: f0 = 5750; Q = 0.707; gain = -3.5

MF shelf:

Low Shelf: f0 = 900; Q = 0.55; gain = -1.0

LF shelf:

Low Shelf: f0 = 250; Q = 0.65; gain = -4.25

* Note that the UHF / distance shelf is a standard filter I use, with varying gain.  My baseline is flat (direct sound) out to 20 kHz at the seats, but for music, I use a default gain = -3.501 (-3 dB @ 20 kHz).  Some slight UHF roll-off is natural even from flat speakers, when listening a few meters away.  Hence, if you already have some roll-off, you may want to attenuate less here, though it probably doesn't make a big difference.  In my experience, changing this gain value has only very subtle effects and you may be able to omit it if short on filter slots.

Comments:

At the time of this post, I've only watched the first 30 minutes or so of the movie.  I'm planning to play it for guests this weekend and don't want to spoil it for myself complete.  I doubt this re-EQ is at all optimal, but I think it makes a big improvement over the baseline.

Without correction, I found the soundtrack to exhibit mostly low frequency bloat.  The excess brightness was very subtle until I started shelving the lows down.  Much of the early dialog consists of women with fairly low voices, with fundamentals landing in the 160-220ish Hz range.  They are often muddy and boomy without the corrections.  Even with corrections, they are very full sounding for me, but I didn't want to shelve away any more bass to try to retain as much punch as possible.  What bass I got to hear sounded (and felt) really good, especially with the solid LF extension.

I was surprised by how much positive difference the MF shelf made.  The extra upper mids really brought out the richness of the vocals and helped tamp down the brightness that was otherwise there with only -2 dB/octave on the HF.  By my last trial through the beginning of the movie, I noticed the sound effects are a lot cleaner and a lot more ambiance is revealed.  The score is more engaging and sounds more 3D.  Some background dialog that occurs shortly after the amazons first appear can be heard clearly whereas before it seemed overpowered by the music.

Comments are welcome, and I'll make an edit note if I update the EQ.

Edit / EQ update: calling this one v2; response shape plot shows comparison with previous

Edit2: I no longer recommend anyone try applying this.

[SME]

Just to clarify once again, these corrections are likely to work best on a system that sounds neutral for music.  This is most easily achieved by using anechoic flat speakers and avoiding EQ except as needed in the low frequencies to knock down peaky resonances and to compensate for baffle-step / boundary effects for full-space speakers placed near walls and/or corners.  Such systems are likely to have an in-room frequency response with a lot more bass (e.g. 6-10 dB higher) than treble rather than the kind of flat-ish response one gets by running one of the Audyssey MultEQ techs.  I need to post a picture of mine, but these days I'm around +9 dB in the bass compared to 4-8 kHz, and I still have some top-end roll-off.

If you are running Audyssey, then I'm sorry, I can't really help you.  IMO, Audyssey is flawed in similar but different ways from X-curve.  It will not deliver consistency between different rooms, placements, and speakers.  Your best hope would be to find a correction for your system that makes it neutral for music and then go from there.  Good luck with that.  If you actually have generic EQ ability, you ought to use it instead of Audyssey anyway, IMO.

Also it's worth pointing out that the black curve in post #2 is not much different from what happens in a typical small listening room, and this is probably not a coincidence.  I believe relative in-room response shape for a "flat" speaker may have more to do with the speaker directivity vs. frequency than anything else.  Room size does still affect relative loudness vs. calibration level, and I suspect that this is because of reverb build-up in the larger and/or livelier room that is mostly independent of frequency and therefore does not appreciably affect the relative response.  My assumption here is that reverb energy inflates the pink noise SPL meter reading without appreciably increasing loudness.  Likewise, I wouldn't be surprised if 85 dBC is just too loud of a calibration target for more modern cinemas with more treatments, excepting the ones with really big rooms.

[minnjd]

Awesome work SME.  My head hurts from reading it but there's a lot of good information.

[Infrasonic]

Oy vey.

 

Harman Target Curve.

[SME]

I think there are a few curves that one could identify as the Harman target curve.  The one I think of as the Harman curve consists of a smooth constant slope, i.e. the red curve in this picture:

I think that's likely to be a big improvement over flat for most people.  However, Harman has since made refinements.  What I see labeled as the Synthesis curve (Synthesis is a brand under Harman) may be a better starting point, albeit with some adjustments noted below:

(1) There's no good reason to roll-off below 25-30 Hz as shown.  The only reason that roll-off is probably there is that the Synthesis subs don't play that low. 

(2) The roll-off shape at the top-end is a lot closer to a distance roll-off, but it doesn't match quite right.  IMO, starting with a flat baseline and applying a proper distance roll-off.  See the high shelf in "UHF / distance" in the "Wonder Woman" EQ config for a very good distance roll-off approximating shape I developed.  Vary the gain as desired.

(3) A system with a curve like this will probably be mid-bass deficient.  With music, I prefer more bass boost above 100 Hz and less in the 40-100 Hz range.  Otherwise there's no bass punch and speed/definition is poor.  It's weird they settled on that curve because I anticipate that a Revel Salon 2 (another Harman product) placed away from walls and run without EQ would likewise measure with more of a boost between 100-200 Hz.

Presumably Harman developed this target in part by measuring one of their flagship speakers without EQ in their test room.  They may have supplemented it with some listener tests, such as the ones where they have subjects raise and lower "treble" and "bass" controls.  I'm not a fan of these latter tests as I'm not convinced they adequately control for reference level differences and the controls are too crude.

Then there's a curve labeled "Synthesis 2":

This one has much less distance roll-off at the top, and the bass boost appears to be centered a bit higher, closer to 90 Hz than in the "Synthesis 1" curve.  I know for a fact that Synthesis introduced a new curve over the last year, but the company is treating it as proprietary information and trying to keep it a secret.  I think that's just silly, especially because I really don't think any of these curves is a perfect match for every system.

As I've mentioned many times, I believe the optimal target curve is dependent on the speaker and on placements / distances and nearby boundary effects, and overall room acoustics.  The method I'm developing and which I use for my system relies on analysis of in-room measurement data, taken at multiple locations, weighted according to distance, and interpreted in the context of the known speaker characteristics.

@Infrasonic, your system would make an interesting case study for my calibration approach because it appears to retain a lot of directivity throughout its range.  I just now took a peak at your FR posted in your AVSForum build thread, and I have to say, they look very close to what I would what have recommend you to try for a neutral sound.  @Infrasonic's response is flat up top with a modest slope over the 2-6k range where the SEOS-12 loses some directivity (vertically).  Then it stays pretty flat all the way down to almost 200 Hz.  This is to be expected because his system probably loses very little directivity going all the way down to the bottom.  His bass boost is centered closer to 150 Hz, which is where mine is and is IMO much better than the Synthesis targets. 

I need to post my updated room curves but I'm waiting until I correct a minor resonance at 500 Hz in my L/R.  It's on my TODO list.

[maxmercy]

SME,

Have you looked at the entire WW soundtrack's PvA full range (20-20) and seen what your changes would look like?  What FFT settings did you choose if you did this?

JSS

[SME]

10 minutes ago, maxmercy said:

SME,

Have you looked at the entire WW soundtrack's PvA full range (20-20) and seen what your changes would look like?  What FFT settings did you choose if you did this?

I have not.  I still don't have a way to get the PvA(s).  Hopefully in a few months I'll get a PC with BDR drive built.

I'm certainly curious as to how it changes and whether one can tell by looking at a PvA whether or not it would benefit from this kind of correction.  However, I kind of doubt it will work as well as for BEQ.  In general, I expect the PvA is dominated by high SPL sounds, especially the peak trace.  The spectral composition of the high SPL sounds probably varies a lot by movie.  The dialog may tend to be a bit more consistent between movies, but I doubt the dialog will dominate the PvA, especially in action movies.  Also note that the adjustments here are quite a bit smaller than for BEQ.

On another note, I updated the EQ on "Wonder Woman" slightly.

[3ll3d00d]

I have wonder woman so can extract this for you with and without your correction on top if you like. Just post what you want to see and I will get to it over the next few days.

[SME]

1 hour ago, 3ll3d00d said:

I have wonder woman so can extract this for you with and without your correction on top if you like. Just post what you want to see and I will get to it over the next few days.

Hey!  Thanks for the offer.  Can you show PvA (both peak and average) of before and after for *just the center channel*?  I developed the correction by ear with insight from the theory described above, and it was the center channel and dialog that most strongly influenced my subjective assessment.  If there's anything worth seeing in the comparison, it should show up in the center channel traces.

Can I ask another favor?  You are one of the few people who have the actual capability to implement this re-EQ.  Can you try out the "Wonder Woman" re-EQ and let me know what you think?  I forgot to mention, but the filter types are the MiniDSP type.  Do you know how to convert them to JRiver?  None of the shelf Qs are more than 0.707, so I think it should be do-able.

[3ll3d00d]

OK, I'll pull that out for you.

I do know how convert to jriver format filters. Are there any particular scenes (or segments of the film) I could use for this?

[SME]

To follow on my last post.  How many people actually have the capability to implement re-EQ like this?  I'm guessing that very few can do so.  Even those with EQ capability probably have very few free slots available after room correction.  The "Wonder Woman" EQ above can be implemented with as few as 4 filters if one omits the two that are least important.  I'm guessing most corrections will require 3 or 4 filters at a minimum.

A tip for MiniDSP users:  The MiniDSP devices I've used allow for "custom biquads" to be input in the crossover section(s).  I've found that this is effectively opens up a lot more filters than one normally has access.  That's because the XO section has enough spare biquads for XOs that are very high order, which most people don't use.  So to access these extra filters, I suggest using a biquad calculator to implement the XO(s) together with as many room/speaker EQ PEQs as you can entirely within the XO "custom biquad" block.  This allows for more of the standard PEQs (in the input or output sections) to be used for EQing content.

Another tip:  There's no need to implement the MF or HF PEQs on the SUB/LFE channel because the filters don't do anything below 120 Hz.  Instead a simple gain adjustment (in the cumulative amount of the low-shelves) should be enough.  This is a real plus being that most people use more filters on their subs for room EQ.  It may be okay to omit the LF shelf on the LFE/SUB channel too, but I would try to include it if possible.  Even 0.25 dB is perceptually significant when looking at coarse / wide-band  response changes.

[SME]

15 minutes ago, 3ll3d00d said:

OK, I'll pull that out for you.

I do know how convert to jriver format filters. Are there any particular scenes (or segments of the film) I could use for this?

Thanks!  Any particular scenes?  I dunno.  I only tested the first 30 minutes.  I'm going to watch the full movie with the re-EQ on Saturday night.

I would say that just about anything in the movie is a fair test.  Unlike BEQ which is most apparent with big effects, this re-EQ will change the sound everywhere and should be especially notable with dialog.  If you can do a quick A/B swap, it's even better.  This is something I should try now that it's "done".  You might want to add some gain to the re-EQ config to compensate for loudness differences so you can swap quickly without having to manually adjust the master volume.

[SME]

Star Trek (2009, BD / Dolby TrueHD 5.1)

UHF / distance (relative to flat response; see note*):

High Shelf: f0 = 13900.0; Q = 0.4675; gain = -6.613

HF slope (see note **):

High Shelf: f0 = 3140; Q = 0.707; gain = -2.0
High Shelf: f0 = 7750; Q = 0.525; gain = -2.0

MF shelf:

Low Shelf: f0 = 625; Q = 0.675; gain = -0.375

LF shelf:

Low Shelf: f0 = 150; Q = 0.625; gain = -2.75

* Note that the UHF / distance shelf is a standard filter I use, with varying gain.  My baseline is flat (direct sound) out to 20 kHz at the seats, but for music, I use a default gain = -3.501 (-3 dB @ 20 kHz).  Some slight UHF roll-off is natural even from flat speakers, when listening a few meters away.  Hence, if you already have some roll-off, you may want to attenuate less here, though it probably doesn't make a big difference.  For this track, I found that things were a tad forward without the extra attenuation, so I don't recommend omitting it.

** These two filters are calculated so as to provide a -2 dB/octave slope down to 8-10 kHz.  The filter to sharpen the 2 kHz knee was omitted because it sounds better without it.

Comments:

The soundtrack to "Star Trek (2009)" is among the most praised and not just for its dynamic, full-bandwidth bass.  However, it is likely a cinema mix that can be improved with re-EQ.  Without correction, this one actually sounds pretty good and is only a little bass heavy.  However, it can be uncomfortably bright at times, especially in some of the louder sequences.

Applying a partial -2 dB/octave slope on the HF helps tame the highs and gives a better sense of depth to the score as well as improving audibility of the mids.  This mix is unusual in that it seemed to need a bass shelf a lot lower than most.  Dropping the sub bass level a bit relative to ~180-200 Hz should help make the bass a bit more punchy.

Where I  struggled was with the mid-range, and I think that there's still room for improvement there.  I suspect that they started boosting the top below 2 kHz, because with this correction, there seems to be some peakiness there.  The MF shelf was kind of a compromise.  To do it right, I'd have to modify the f0 and Q for first shelf used for the HF slope, and that would complicate my implementation.  This is probably something I should do anyway, being that a lot of other mixes will likely have this kind of problem.  As such, I'll likely make more improvements to this one later.

Comments are welcome, and I'll make an edit note if I update the EQ.

 

Edit: I no longer recommend applying this EQ.

[3ll3d00d]

This is the before C channel for Wonder Woman, I ran it through speclab with no decimation and 4096 fft length to give ~11Hz frequency resolution

https://imgur.com/gIHjYdF

 

[maxmercy]

What window did you use? 

It seems that to get a 'real' picture of what is happening with a PvA for this type of correction, we would need a reference 'line' placed on the avg graph or tilt the average graph counter-clockwise to see deviations from 'level' (a downward-sloping target curve).

Main thing I see here are the broad 'boosts' centered around 1kHz and 8kHz.

WaveShop is probably a better app to use for these Spectrums, and very fast, too. 

Interested to see the 'after', and if you can, do a pre-post image shifting the colors for the post.

JSS

[3ll3d00d]

Nuttall

I made the  image a bit bigger and included pink noise spectrum as a reference curve for comparison - https://imgur.com/a/yxK3V

If there is some other reference curve that would be better then let me know & I can update it. I'll get to the after pic tonight probably.

[3ll3d00d]

 

Wonder Woman (2017, BD / Atmos) - JRiver equivalent filters

NB: I refer to it as S below to differentiate from the original post, jriver calls this Q though.

measurement to verify shape -> https://imgur.com/a/05Ixe

UHF / distance (relative to flat response; see note*):

High Shelf: f0 = 13900.0; S = 0.442082; gain = -3.501

HF slope (see note **):

High Shelf: f0 = 3140; S = 0.9997; gain = -2.0
High Shelf: f0 = 7750; S = 0.552885; gain = -2.0
EQ Peak: f0 = 1950; Q = 1.75; gain = +0.33

MF shelf:

Low Shelf: f0 = 1000; S = 0.720521; gain = -1.25

LF shelf:

Low Shelf: f0 = 265; S = 0.785708; gain = -4.0

[3ll3d00d]

Wonder Woman - corrected vs pink noise - https://imgur.com/a/KptYn

[3ll3d00d]

Wonder Woman - corrected vs uncorrected - https://imgur.com/veQRiUG

[SME]

Wow, thanks!  This data is interesting, but it's not clear how best to use it.  I gather that much of the problem arises due to the fact that the center channel incorporates a variety of different kinds of effects.  The approximate slope which sounds most natural may depend on the type of effect.  For the "corrected" version, I have noted two different regimes with different slopes:

It's not at all clear from the images whether my adjustments in the low frequency regime provide an improvement.   The most obvious effect of my "corrections" is the change in slope.  What slope is correct?  Can we assume that the spectral average is dominated by dialog?  Or could the score and/or effects also significantly influence this average?  If so, what slope is natural sounding for dialog?  For the score?  For effects?  Are these slopes even consistent for the same types of sounds?

This M.S. thesis studying spectral slope indicates that many (if not most) musical instruments exhibit different slopes depending on how loudly they are played and what fundamental frequency is involved.  Furthermore, the slopes can change as the note decays.  Here's an interesting article covering the problem of voice intelligibility in detail and includes typical spectra of speech for different kinds of human speakers and different manners of speaking (i.e. casual voice to shouting).  What I take away from all this is that we really can't re-EQ the full-range track based on the PvA data.  Doing the job right requires interpretation of the sound in the context of what's happening on screen, and without some kind of clever A.I. solution, this requires a human interpretation.

With that said, it's worth noting that the "corrected" curve in the high frequency regime, above 800 Hz, appears to follow a constant slope a bit more closely.  This begs the question of whether the correction could be improved further by shelving less over 2-4kHz and shelving more over 4-8 kHz.  I may have to give it a try.

[SME]

So I decided to break with the constant X-curve slope in the HFs and see what happens if I allow a bit more upper-mid and a bit less treble in the "Wonder Woman" EQ config.  The biggest change is the boost to 2.5-5 kHz.  That netted an improvement, so I went ahead and worked on smoothing out the rest of the mids a bit more.

This config sounds very sweet at "-6" for me and uses one fewer filter, the PEQ for the X-curve knee, which wasn't helping.  Is it perfect?  No.  Nothing's perfect.  :)  I'll listen carefully while I watch the entire movie Saturday night before deciding whether to try to improve on it.  From what I am hearing (and my biased opinion), I think this re-master could easily compete with the better home mixes, e.g. from Disney.

[SME]

We watched "Wonder Woman" tonight.  I made some additional tweaks, primarily filling out the low frequencies a bit more (albeit a bit too much, in retrospect) and cutting back a bit more at and below 120 Hz to keep the low voices from booming too much.  It sounded quite good, but was a tad excessive in the 150-300 Hz area.  I want to try to improve on that a bit further before posting another EQ configuration, and I may also try to bring out a little more top-end.  Other than that's it's sounding very good.

The movie and its sound impressed me a lot.  The beginning of the movie was a bit on the loud side, so I was surprised when it switched gears mid-way through and got significantly louder still.  The re-master really shined here, which is a very good sign.  Imbalances in the mids and highs tend to reveal themselves more by causing discomfort when stuff gets loud.  This presentation was very smooth and very big.  I loved the use of bass in this one, and I noted multiple effects, both transient and continuous, that were very impressive.  The effects were varied and used the different frequency ranges well, including infrasonic.  Though, it's possible it could benefit from a bit more infrasonic sound, especially given the extra headroom available after the re-master.  Any thoughts on that @maxmercy?

Anyway, I'll try to get around to "truly finishing" the WW re-master, maybe early next week.  This movie is definitely a good demo movie, both for my system and for the meta-data re-mastering concept.