• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Room/Speaker EQ Approaches: Parametric EQ settings vs Impulse/Convolution

LightninBoy

Senior Member
Forum Donor
Joined
Jan 9, 2019
Messages
481
Likes
825
Location
St. Paul, MN
Hi folks, I'm going to jump right to heart of the question. See below *Background* section for further background and context.

In my research, I see two general approaches to taking room measurements in the context of a near-field room and speaker correction. And a few variations within:

1. Take measurements of all speakers at and around the main listening position. Use REW to produce EQ settings, export into text file and import into EQ software like Equalizer APO to apply corrections. . The eq settings consist of Freq, gain, and Q.

1 (variation). Take individual measurements of speakers (left and right). Use REW to produce EQ settings and export into text file. Import into EQ hardwares/software that supports individual L/R EQs.

2. Take individual measurements of speakers (left and right). Use REW to produce an Impulse response file for each speaker. Import those into a convolver plugin to apply corrections.

So my question is, what are the *practical* differences between these approaches? Why/when would I do one or the other? Is one more accurate? Does one allow you to do more equalization with less artifacts?

For example - one thing I understand about approach #2 is that your IR files must be generated for a specific sampling rate. And when listening back, you need to use the IR file for the sampling rate that is playing. This is a clear practical difference between approach #1 vs #2. What are some others?

** Background **
I've been digging into the measurement and EQ topics for a few weeks now. I'm not exactly a newbie - I did started working with REW many years ago to help set up the sub in my HT room. But much has changed so it took me a long time to weed through this all. For my recent measurements, I just used a mic I had on hand to get familiar with the software and process again. But my UMIK came today, so its time to get serious.

My goal this time around is to measure and correct the speakers in my "recording studio". So the primary goal is for a flat response for mixing music. However, I do find myself listening to more music on this system, so simply enjoying music playback is a goal as well. Some equipment/software currently in use ...
* Speakers - JBL 306mkII
* Sub - Tannoy TS10
* Interface - Behringer UMC404HD
* DAW - Reaper
* Music Playback - Foobar (no Roon or JRiver)
* Other software: Equalizer APO and many many plugins for Reaper
 

vavan

Senior Member
Joined
Apr 29, 2019
Messages
341
Likes
201
Location
Kazan, Russia
one thing I understand about approach #2 is that your IR files must be generated for a specific sampling rate. And when listening back, you need to use the IR file for the sampling rate that is playing
I believe some convolvers can automatically resample IRs for different sample rate inputs
 

hyperplanar

Active Member
Joined
Jan 31, 2020
Messages
299
Likes
531
Location
Los Angeles
Option 2 is the worst as you’re only correcting what your mic heard from a single point in space. You should do a sweep measurement as well as MMM RTA of L, R, and L+R. The sweeps will give you insight as to what response deviations are minimum phase or not (as I understand it, these are readily correctable while excess phase is usually not). Then once you have an idea of what you’re dealing with, use an appropriately sloped target response on the MMM measurements to obtain the parametric EQ settings needed for correction.

It makes sense most of the time to use minimum phase EQ for room correction (meaning a normal run-of-the-mill EQ, not a linear phase one). This is because room modes are minimum phase phenomena, meaning that if you correct their frequency response with a minimum phase EQ, you also automatically correct their phase response at the same time, which kills their ringing as well.

Linear phase filters (which are usually what IR convolution is deployed for) can be used in more specific situations, a good one is for crossover phase linearization between the woofers and tweeters, because you can do this with a latency of ~1 ms and the frequency is high enough to avoid potentially audible preringing. The subwoofer crossover phase shift is not correctable without a lot more latency (which is probably unacceptable for your use), and may also introduce audible preringing.

By the way, without a hardware DSP I think you’ll find doing room correction to be an annoyance. EqualizerAPO doesn’t work for ASIO clients like your DAW, so you’ll have to end up sticking an EQ plugin on your master or monitor bus for every single project you do.
 
Last edited:

QMuse

Major Contributor
Joined
Feb 20, 2020
Messages
3,124
Likes
2,649
2. Take individual measurements of speakers (left and right). Use REW to produce an Impulse response file for each speaker. Import those into a convolver plugin to apply corrections.

This wouldn't work. Convolver needs something like inverted IR to get flat response, not the actual IR of the speaker. That is simple to do but it would sound awfull. What you need is correction based on spatially averaged mesaurements adjusted to your preferred tonal balance. Once you got FR to be as you like it you can measure IR and get phase correction with a tool like rePhase, or you can use automatic tool like Dirac which would do all of that for you.
 

QMuse

Major Contributor
Joined
Feb 20, 2020
Messages
3,124
Likes
2,649
The subwoofer crossover phase shift is not correctable without a lot more latency (which is probably unacceptable for your use), and may also introduce audible preringing.

Why would latency be a problem with music listening? And even if XO with sub and mains is implemented via AV processor they have video delay feautre to fix lipsync with the delay introduced to mains.
 

hyperplanar

Active Member
Joined
Jan 31, 2020
Messages
299
Likes
531
Location
Los Angeles
Why would latency be a problem with music listening? And even if XO with sub and mains is implemented via AV processor they have video delay feautre to fix lipsync with the delay introduced to mains.
My goal this time around is to measure and correct the speakers in my "recording studio".
Gonna have a hell of a time monitoring stuff once the delay passes more than 10 ms or so... :)
 

digitalfrost

Addicted to Fun and Learning
Forum Donor
Joined
Jul 22, 2018
Messages
794
Likes
1,254
Location
Palatinate, Germany
If all you do is create an EQ, IMHO it does not matter wether you apply that with a parametric EQ or a convolver. However, convolution can do more than just EQ, like correct the time domain, and do frequency dependent windowing (FDW) i.e. less correction towards the higher frequenices.

So convolution can be much more than just an EQ. I don't think there will be a difference between your solutions 1 and 2, given that REW will produce identical output.

I would opt for a convolver solution but produce correction files with DRC-FIR, Acourate or another solution.
 

QMuse

Major Contributor
Joined
Feb 20, 2020
Messages
3,124
Likes
2,649
However, convolution can do more than just EQ, like correct the time domain, and do frequency dependent windowing (FDW) i.e. less correction towards the higher frequenices.

FDW is usually applied in the filter creation phase, not in the convolution engine which only applies filter to the input signal.
 
OP
LightninBoy

LightninBoy

Senior Member
Forum Donor
Joined
Jan 9, 2019
Messages
481
Likes
825
Location
St. Paul, MN
This wouldn't work. Convolver needs something like inverted IR to get flat response, not the actual IR of the speaker.

I probably didn't describe the steps well, but that is essentially what REW does if I understand correctly - generate an impulse response that represents the EQ filters. Here are the steps as described by Roon ...

https://blog.roonlabs.com/digital-room-correction/

In any event, this approach is documented in several places so I'm sure it works. What I am trying to determine is how is it different than simply applying the eq via a parametric EQ plugin.

That is simple to do but it would sound awfull. What you need is correction based on spatially averaged mesaurements adjusted to your preferred tonal balance. Once you got FR to be as you like it you can measure IR and get phase correction with a tool like rePhase, or you can use automatic tool like Dirac which would do all of that for you.

Lets focus just on FR for now, in other words, how to adjust the sound to my preferred tonal balance. For this, when would one use the IR/Convolver approach to fix FR vs the parametric EQ approach? Are you saying never - it provides no benefits over parametric EQ settings? If so, why does the Roon article above use the IR/Convolver approach then vs a much simpler parametric EQ approach?

Appreciate everyone's help on this.
 
OP
LightninBoy

LightninBoy

Senior Member
Forum Donor
Joined
Jan 9, 2019
Messages
481
Likes
825
Location
St. Paul, MN
Option 2 is the worst as you’re only correcting what your mic heard from a single point in space. You should do a sweep measurement as well as MMM RTA of L, R, and L+R. The sweeps will give you insight as to what response deviations are minimum phase or not (as I understand it, these are readily correctable while excess phase is usually not). Then once you have an idea of what you’re dealing with, use an appropriately sloped target response on the MMM measurements to obtain the parametric EQ settings needed for correction.

It makes sense most of the time to use minimum phase EQ for room correction (meaning a normal run-of-the-mill EQ, not a linear phase one). This is because room modes are minimum phase phenomena, meaning that if you correct their frequency response with a minimum phase EQ, you also automatically correct their phase response at the same time, which kills their ringing as well.

Linear phase filters (which are usually what IR convolution is deployed for) can be used in more specific situations, a good one is for crossover phase linearization between the woofers and tweeters, because you can do this with a latency of ~1 ms and the frequency is high enough to avoid potentially audible preringing. The subwoofer crossover phase shift is not correctable without a lot more latency (which is probably unacceptable for your use), and may also introduce audible preringing.


Many thanks for the detailed response.

You introduced some new terminology so let me make sure I'm tying this back to my original question correctly. I previously noted two different approaches widely discussed for room/speaker correction with REW ...

1. Take measurements of all speakers at and around the main listening position. Use REW to produce EQ settings, export into text file and import into EQ software like Equalizer APO to apply corrections. . The eq settings consist of Freq, gain, and Q.

2. Take individual measurements of speakers (left and right). Use REW to produce an Impulse response file for each speaker. Import those into a convolver plugin to apply corrections.

So here is what I takeaway from your post ... When you say "minimum phase EQ" this relates to standard parametric EQ settings as one would get when exporting the REW EQ filters as text and applying those in a parametric EQ plugin. So approach #1 is using minimum phase EQ.

In the context of apporach #2, when REW creates the IR file based on the calculated EQ settings - it is basically just applying these minimum phase EQ filters. So approach #2 is also using minimum phase EQ filters. So there's no practical advantage to this approach. Although IR/Convolvers are capable of more sophisticated corrections such as "linear phase filters", REW is not incorporating those into the IR file.

If that is the case, then I'm really struggling to understand why anyone would use or recommend approach #2.

By the way, without a hardware DSP I think you’ll find doing room correction to be an annoyance. EqualizerAPO doesn’t work for ASIO clients like your DAW, so you’ll have to end up sticking an EQ plugin on your master or monitor bus for every single project you do.

Yep, understood. I may end up going with a hardware solution.
 
Last edited:

QMuse

Major Contributor
Joined
Feb 20, 2020
Messages
3,124
Likes
2,649
I probably didn't describe the steps well, but that is essentially what REW does if I understand correctly - generate an impulse response that represents the EQ filters. Here are the steps as described by Roon ...

Sure, REW can generate EQ filter, but your wording was a little off so it could lead to wrong conclusions.
 

txbdan

Active Member
Joined
Apr 21, 2020
Messages
157
Likes
140
I'm also interested in this topic. I've used REW to create a set of parametric EQ settings against the target response and applied them. The results weren't very good.

I then tried using IK ARC3 and the results weren't very good.

Then i tried the Dirac Live 2.0 demo and holy crap its excellent. Of course Dirac has their proprietary magic, but if i could somehow recreate something close to this using free or cheaper tools like REW i'd be quite the happy camper. But assuming i can't, i might actually be willing to pony up $350 for a simple 2.1 office music setup. It's just so good. OP, you may want to give the DL 2.0 demo a try as a data point.
 

hyperplanar

Active Member
Joined
Jan 31, 2020
Messages
299
Likes
531
Location
Los Angeles
So here is what I takeaway from your post ... When you say "minimum phase EQ" this relates to standard parametric EQ settings as one would get when exporting the REW EQ filters as text and applying those in a parametric EQ plugin. So approach #1 is using minimum phase EQ.
Yup, that's right. Any standard zero-latency parametric EQ, digital or analog, is minimum phase. EqualizerAPO, your DAW's channel EQ, etc.

In the context of apporach #2, when REW creates the IR file based on the calculated EQ settings - it is basically just applying these minimum phase EQ filters. So approach #2 is also using minimum phase EQ filters. So there's no practical advantage to this approach. Although IR/Convolvers are capable of more sophisticated corrections such as "linear phase filters", REW is not incorporating those into the IR file.

If that is the case, then I'm really struggling to understand why anyone would use or recommend approach #2.
I've never tried using REW itself to generate a correction IR, but assuming it does generate a minimum phase EQ in the process, this is probably just to allow people to have a bit more flexibility.
  • For one, some parametric EQs don't behave the "textbook" way with the gains and Q's put into them. FabFilter Pro-Q, for instance, where a Q of 1.00 is equivalent to a Q of 0.707 in most other EQs. So when copying in the parametric EQ settings that REW spits out, there is a potential mistake to be made here in that your EQ plugin doesn't follow the same conventions as REW. It can be much easier to just copy a .wav file to get your filters into a system like Roon, and guarantees that the shape of the filter is exactly the same as REW calculated.
  • Many hardware DSPs have a limited number of parametric EQ filters available, while they also offer IR convolution. You can incorporate a basically unlimited number of parametric EQ filters into a correction IR, and then just use the IR convolution instead to get around this limitation.
The main con is that IR convolution is more resource-intensive than the standard parametric EQ IIR filters, and this gets worse the lower you go in the frequency range. Given that you have something such as EqualizerAPO or a miniDSP, I would prefer to load up the parametric EQ settings in those instead of doing convolution. You can use convolution as the final touch to linearize the crossover phase shift at a later point—IMO the room correction is waaaay more important.

Yep, understood. I may end up going with a hardware solution.
Assuming the DSP has more than 2 outputs, this would also allow you to correctly integrate your sub by allowing you to EQ the sub and speakers separately, as well as setting an appropriate delay to let the two sum as constructively as possible. Considering that, I think it's the only way to go!
 
Last edited:

Wayne A. Pflughaupt

Active Member
Joined
Mar 14, 2016
Messages
285
Likes
230
Location
Corpus Christi, TX
I'm also interested in this topic. I've used REW to create a set of parametric EQ settings against the target response and applied them. The results weren't very good.
Manual EQ can be tricky. You have to know what should be EQ’d and what should be left alone. Then you have to know what can be EQ’d and what can’t.

A common mistake is going overboard and plastering the whole curve with a bunch of filters that aren’t even going to make an audible difference.

In addition, care should be taken applying different filters to the main L/R speakers. Mismatched filters above ~300 Hz can whack the imaging. Only below that point should mis-matched filters be attempted.

Regards,
Wayne A. Pflughaupt
 
Top Bottom