The audible differences with Equilibrium made me think about the tab count and shape of the IR.
Equilibrium shows it like this,
8K tab count (impulse file length 16K)
And 65K tab count, (impulse file length 131K)
This does make me think again about pré-ringing and it's audibility. I didn't realize even the length of the IR for FIR filters was a factor in this.
Not that I need to worry with a 65K tab count in my own processing. So besides windowing, even the tab count could influence things like transient perception. Further depending on the steepness of the filters used of course. But I've never seen it pictured like this before.
For processing in my current chain I use 16K tab count (32K file length) within Equilibrium and the above pictured setting for the Kaiser window.
It's the current sweet spot between sweet sound and responsiveness of the JRiver system with 8 instances of Equilibrium running in total.
4 in Linear phase FIR, 2 in Analog FIR and 2 in Digital + Compensation and Phase. That keeps everything responsive, the higher the setting the less responsive it feels. Home Theatre is currently running one step lower in FIR file length. Next up I need to test it long run.
Equilibrium shows it like this,
8K tab count (impulse file length 16K)
And 65K tab count, (impulse file length 131K)
This does make me think again about pré-ringing and it's audibility. I didn't realize even the length of the IR for FIR filters was a factor in this.
Not that I need to worry with a 65K tab count in my own processing. So besides windowing, even the tab count could influence things like transient perception. Further depending on the steepness of the filters used of course. But I've never seen it pictured like this before.
For processing in my current chain I use 16K tab count (32K file length) within Equilibrium and the above pictured setting for the Kaiser window.
It's the current sweet spot between sweet sound and responsiveness of the JRiver system with 8 instances of Equilibrium running in total.
4 in Linear phase FIR, 2 in Analog FIR and 2 in Digital + Compensation and Phase. That keeps everything responsive, the higher the setting the less responsive it feels. Home Theatre is currently running one step lower in FIR file length. Next up I need to test it long run.
Thanks for tagging me as I had not been getting notifications for the thread. I did send you Equilibrium in October 2017, so I'm glad you finally like it
It is still the only EQ that behaves closely enough to PL par EQ on your old Mid Side settings to replace it.
It took me long enough! (lol) I waited till I finally changed to Windows 10 Pro. Still not liking that move though, but I had to do it at some point.
So what do you think about those graphs as presented by Equilibrium based on FIR file length.
If I compare an impulse created with 32K FIR length and one created with 16K FIR length (as saved to disk) within REW and in Filtered IR view I see no difference whatsoever. I took one of the nastiest EQ signals I had for that comparison. I'll try it again tonight with a bigger difference, but I won't hold my breath.
So what do you think about those graphs as presented by Equilibrium based on FIR file length.
If I compare an impulse created with 32K FIR length and one created with 16K FIR length (as saved to disk) within REW and in Filtered IR view I see no difference whatsoever. I took one of the nastiest EQ signals I had for that comparison. I'll try it again tonight with a bigger difference, but I won't hold my breath.
While there were minute differences to be seen between the files, an 8K and a 32K filter are exactly the same upon first gaze. The difference (in this case) was the reverb, which is different for every time an IR flows trough it. If I subtract the files with reverb, part of the reverb turns up as a difference signal.
If I leave the reverb off and subtract the 8K file from it's 32K counter part I get a difference file that is 57.8 dB lower in level at it's highest point (70 Hz). That could be audible.
Level of both the files that were subtracted from each other was 108.1 dB at 70 Hz.
All visual graphs seem the same though, so it's going to be hard to tell. So use what seems appropriate and "sounds good to you".
Subtracting two files generated with FIR files of the same length do null out.
I used the 32 bit version of JRiver to run these tests, I doubt there is a difference between that version and the 64 bit version as both use internal 64 bit processing within the DSP engine.
Take this for what it is, but I am a bit skeptical about the graphs within Equilibrium that show differences in filter ringing between files with different impulse length. I'd say if that were true, a lot more Pro gear would feature longer FIR file lengths to be able to circumvent this.
If one uses steep filters, like brick wall or deep surgical cuts with high Q, there may be differences. For the general use like I do here (gentle EQ shaping and slow roll off filters) I kind of doubt I could hear it. I'll try in next listening session. I bet I save a lot of processing power with shorter FIR file lengths for these simple jobs, but when in doubt (about audibility) I can revert to using 32K length files. I had set my Home Theatre correction to 16K, but after this test I definitely will try 8K.
If I leave the reverb off and subtract the 8K file from it's 32K counter part I get a difference file that is 57.8 dB lower in level at it's highest point (70 Hz). That could be audible.
Level of both the files that were subtracted from each other was 108.1 dB at 70 Hz.
All visual graphs seem the same though, so it's going to be hard to tell. So use what seems appropriate and "sounds good to you".
Subtracting two files generated with FIR files of the same length do null out.
I used the 32 bit version of JRiver to run these tests, I doubt there is a difference between that version and the 64 bit version as both use internal 64 bit processing within the DSP engine.
Take this for what it is, but I am a bit skeptical about the graphs within Equilibrium that show differences in filter ringing between files with different impulse length. I'd say if that were true, a lot more Pro gear would feature longer FIR file lengths to be able to circumvent this.
If one uses steep filters, like brick wall or deep surgical cuts with high Q, there may be differences. For the general use like I do here (gentle EQ shaping and slow roll off filters) I kind of doubt I could hear it. I'll try in next listening session. I bet I save a lot of processing power with shorter FIR file lengths for these simple jobs, but when in doubt (about audibility) I can revert to using 32K length files. I had set my Home Theatre correction to 16K, but after this test I definitely will try 8K.
I don't think it's anything to be concerned over, the IR is what it has to be to have the response it does. If you zoomed in enough you would see the same thing with the higher tap count. If you view an IR in REW in dBFS you see the same "ringing" look, viewed as percentage it all goes away
Listening tests seemed to contradict my conclusions above. But then again, it was a flawed test where I enjoyed the JRiver 64 bit sound the most.
Afterwards I did notice some differences and even errors in some of the DSP settings. I need way more time to set it up properly.
The good news, now that I removed all of the Fabfilter instances from all zones, the settings seem to be much more stable, even in 64 bit. I want to be able to use that 64 bit version going forward, as I think there will come a time the 32 bit environment will fall behind in development etc.
I've ordered an M2 NVMe harddrive to use for my O.S. as this type of drive works fine with Win 10 and my trusty (but older) HP Z440 workhorse.
(I would not have been able to boot from a drive like that in Win 7)
For now I'll check the chain after each change in settings, until it becomes clear it's no longer necessary. In hindsight, I think I've been bitten by this bug before. Even in 32 bit JRiver.
But then again, it was a flawed test where I enjoyed the JRiver 64 bit sound the most.Afterwards I did notice some differences and even errors in some of the DSP settings. I need way more time to set it up properly.
The good news, now that I removed all of the Fabfilter instances from all zones, the settings seem to be much more stable, even in 64 bit. I want to be able to use that 64 bit version going forward, as I think there will come a time the 32 bit environment will fall behind in development etc.
I've ordered an M2 NVMe harddrive to use for my O.S. as this type of drive works fine with Win 10 and my trusty (but older) HP Z440 workhorse.
(I would not have been able to boot from a drive like that in Win 7)
For now I'll check the chain after each change in settings, until it becomes clear it's no longer necessary. In hindsight, I think I've been bitten by this bug before. Even in 32 bit JRiver.
These are just visuals from within Equilibrium, based upon the specific EQ settings. There's nothing in it from a measurement or something like that.
So anyone could see how Equilibrium shows these, based on their own EQ settings. The EQ determines the ringing pictured.
But to answer the question: I did not look at the frequency the ringing applies to. As it would still be convolved together with my true sound stream. And that's the combination I get to hear.
Looking at the EQ data alone with a Dirac pulse running trough it while using different setting within Equilibrium shows no differences in REW. So there's nothing from my loudspeaker or even any other audio part in that chain, just a Dirac pulse and Equilibrium (linear phase) EQ.
Dirac pulse in -> JRiver -> Equilibrium -> output write to disk.
I merely showed the DSP windows from within Equilibrium, those show a level difference of ringing due to changing filter length, something I don't see when two different filter lengths are recorded to disk and shown by REW.
But... do I hear an improvement?
So anyone could see how Equilibrium shows these, based on their own EQ settings. The EQ determines the ringing pictured.
But to answer the question: I did not look at the frequency the ringing applies to. As it would still be convolved together with my true sound stream. And that's the combination I get to hear.
Looking at the EQ data alone with a Dirac pulse running trough it while using different setting within Equilibrium shows no differences in REW. So there's nothing from my loudspeaker or even any other audio part in that chain, just a Dirac pulse and Equilibrium (linear phase) EQ.
Dirac pulse in -> JRiver -> Equilibrium -> output write to disk.
I merely showed the DSP windows from within Equilibrium, those show a level difference of ringing due to changing filter length, something I don't see when two different filter lengths are recorded to disk and shown by REW.
But... do I hear an improvement?
Not really to be honest. I'm liking the use of JRiver for it's ease in which I can do whatever I want, and still have a fully functional mediaplayer at my finger tips plus a fully functional Home Theatre setup. I use it as a means to an end. The VST host is the 'engine' that I use to do my own stuff. I try to stick with standard solutions (plugins) and put together something that works for me. I've been using it since 2012, so it probably kind of grew on me I guess. It's kind of like my personal DAW. I never gave up on REW either. I feel I grew up with it.
I've tried using foobar long ago but that didn't last long. Jriver can handle standard VST plugins. I have a lot of flexibility in this setup, so I can change things separately.
I could use it way more efficient by putting together the separate pré EQ I use with the convolution and the room EQ. That would save me 2 instances of Equilibrium right there. But having it setup like this, I can easily redo all separate steps without having to redo everything. Like swapping out the EQ VST or redoing measurements. Or play with tonal balance of the 'in-room' EQ. I kind of use it like an empty shell that I fill with the tools I need. I'm using less and less of JRiver's own processing chain. Running into a space problem after 10 years of use should not be a problem. After a clean up of zones I use, I'm back in business. The most important tools I use are it's convolution engine and the VST hosting. I don't quite see right now how Camilla could be a viable substitute for all of what I do within JRiver. As in Audio player, Home Theatre player with audio/video sync, hosting all the crazy stuff I come up with etc.
A tool/plugin like Metaplugin gives me endless freedom. Unless I do ask too much of it.
I've tried using foobar long ago but that didn't last long. Jriver can handle standard VST plugins. I have a lot of flexibility in this setup, so I can change things separately.
I could use it way more efficient by putting together the separate pré EQ I use with the convolution and the room EQ. That would save me 2 instances of Equilibrium right there. But having it setup like this, I can easily redo all separate steps without having to redo everything. Like swapping out the EQ VST or redoing measurements. Or play with tonal balance of the 'in-room' EQ. I kind of use it like an empty shell that I fill with the tools I need. I'm using less and less of JRiver's own processing chain. Running into a space problem after 10 years of use should not be a problem. After a clean up of zones I use, I'm back in business. The most important tools I use are it's convolution engine and the VST hosting. I don't quite see right now how Camilla could be a viable substitute for all of what I do within JRiver. As in Audio player, Home Theatre player with audio/video sync, hosting all the crazy stuff I come up with etc.
A tool/plugin like Metaplugin gives me endless freedom. Unless I do ask too much of it.
For a similar reason I use Wordpress for my website needs. Yes I've done it in html and java/php and what not. But if I come up with an idea I can almost guarantee that someone already made a plugin that get's me close. I'll simply have to tweak it a little bit to get it to do what I want.
JRiver is like that for me. There isn't an experiment I came up with that I wasn't able to do with it. It has all the flexibility I could wish for.
JRiver is like that for me. There isn't an experiment I came up with that I wasn't able to do with it. It has all the flexibility I could wish for.
Here I am, talking about possible nuances in perception of filter lengths or different Equalizer plugins. Let's see what other advances have been made since the start of this project. Let's compare IR's recorded at (or soon after) the start of this long winded project compared to what it is today.
As measured in Januari 2015 (notes say FIR correction was applied):
If you look closely, you'll see the Stereo measurements shows a double peak at the start, so, worst example.
If I sum the Left and Right speaker from that same session, I get this:
Much better! Which is what I had in Januari 2015. Rough around the edges, no damping panels yet.
So what changed in all these years? I'd say incremental improvements on just about any part of it I could think of.
It made the summed IR look like this (actual measurement at the same spot as the above):
That's progress. A strong reduction in room influence at the listening spot. Measurable and visible this way.
A true measured result of the Stereo pair FIR corrected.
So that's what I got by sticking to one single type of speaker and try to optimize it's use. One more note: these are measured at listening distance, not up close at 1m or less! That makes quite a difference, most measured results we see are not at the listening spot, these are.
As measured in Januari 2015 (notes say FIR correction was applied):
If you look closely, you'll see the Stereo measurements shows a double peak at the start, so, worst example.
If I sum the Left and Right speaker from that same session, I get this:
Much better! Which is what I had in Januari 2015. Rough around the edges, no damping panels yet.
So what changed in all these years? I'd say incremental improvements on just about any part of it I could think of.
It made the summed IR look like this (actual measurement at the same spot as the above):
That's progress. A strong reduction in room influence at the listening spot. Measurable and visible this way.
A true measured result of the Stereo pair FIR corrected.
So that's what I got by sticking to one single type of speaker and try to optimize it's use. One more note: these are measured at listening distance, not up close at 1m or less! That makes quite a difference, most measured results we see are not at the listening spot, these are.
Last edited:
Nope, both of these are single point measurements and corrections based upon single point measurements. In all honesty, with these arrays the single point correction works very well to begin with (for several reasons).
Single point correction always sounded more defined/sweeter to me in the true sweet spot. I've learned what I need to do to make it work more universal (good sounding everywhere) but it simply isn't rewarding (for me) to use (the sum of) multiple measurements with these speakers. In fact, it has lead me to less than stellar results at times with strange stuff happening at (the summed) high frequencies. I'm talking about arrays here in a room that already has room treatment. If you do have (early) reflections, like what I started out with, it would/could be a good idea to try multi point corrections.
If I were to use multi point these days, I'd first measure at a single spot, correct for that and then average multiple points to find differences by applying the correction for that single point to it. Or something like that. Correct for what's different in a larger area. But a vector sum of multiple measurements may not behave like a minimum phase IR.
I used to correct the IR with a mixed phase template in DRC-FIR. These days I correct them with a minimum phase template. The phase edits I want to apply are done manually within RePhase, still using the frequency dependent window techniques. Either the Frequency Dependent Window from DRC or REW (there is a difference between those techniques).
I'm always long winded in my answers, right? Forgive me.
Single point correction always sounded more defined/sweeter to me in the true sweet spot. I've learned what I need to do to make it work more universal (good sounding everywhere) but it simply isn't rewarding (for me) to use (the sum of) multiple measurements with these speakers. In fact, it has lead me to less than stellar results at times with strange stuff happening at (the summed) high frequencies. I'm talking about arrays here in a room that already has room treatment. If you do have (early) reflections, like what I started out with, it would/could be a good idea to try multi point corrections.
If I were to use multi point these days, I'd first measure at a single spot, correct for that and then average multiple points to find differences by applying the correction for that single point to it. Or something like that. Correct for what's different in a larger area. But a vector sum of multiple measurements may not behave like a minimum phase IR.
I used to correct the IR with a mixed phase template in DRC-FIR. These days I correct them with a minimum phase template. The phase edits I want to apply are done manually within RePhase, still using the frequency dependent window techniques. Either the Frequency Dependent Window from DRC or REW (there is a difference between those techniques).
I'm always long winded in my answers, right? Forgive me.
What I like about the arrays is that basically hardly anyone would think it's even possible to get that kind of result when looking at my setup.
After seeing that many drivers, most that see it predict something else entirely. They expect all kinds of things, but not a reduced room influence.
A back wall like you have there is tricky... I'd probably try and get a diffracting pattern on there posing as art.
After seeing that many drivers, most that see it predict something else entirely. They expect all kinds of things, but not a reduced room influence.
A back wall like you have there is tricky... I'd probably try and get a diffracting pattern on there posing as art.