Comparison of DRCs: Dirac Live for Studio, IK Multimedia ARC System 3 and Sonarworks Reference 4 Studio edition

[dominikz]

That is a massive suck out between 50-100Hz, I don't think that can really be cured by any EQ....I think the solution would involve room sound treatments or rearranging speaker & listening position in the room.

You are right - that would be the best solution. However it is not an option for me at the moment.
As it is, EQ still transforms the sound from unlistenable to quite useable and enjoyable. A lot of the energy in the suck-out area is restored (but not all, of course).
Much better than nothing, in any case!

 

[Rottmannash]

Keep in mind that those impulses are loopback impulses, not measured from the speakers. Speakers are typically a minimum phase system, and it looks here like Dirac may be trying to shift the speakers towards more of a linear phase response (hard to tell though what it's doing without impulse response measurements in-room).

This may be way off topic but can someone explain what Dirac does related to phase response?

 

[dominikz]

This may be way off topic but can someone explain what Dirac does related to phase response?

To my understanding Dirac tries to do two things in the time-domain:
1) Adjust channel delay so that sound from all speakers arrives to the listening spot (MLP) at the same time
2) Correct individual speaker's phase response to compensate for the phase shift introduced by crossover circuit and possible misalignment of drivers. The result is an impulse and step response that looks closer to a theoretical ideal.

The correction from 1) above could be beneficial in case of differing distances between individual speakers and MLP (especially with multi-channel, I expect), but seems that research is fairly sceptical of audible benefits of correction from 2) in realistic (non-anechoic) rooms and with real music signals.
IMHO this is an interesting thread and article on the topic.

 

[Robbo99999]

To my understanding Dirac tries to do two things in the time-domain:
1) Adjust channel delay so that sound from all speakers arrives to the listening spot (MLP) at the same time
2) Correct individual speaker's phase response to compensate for the phase shift introduced by crossover circuit and possible misalignment of drivers. The result is an impulse and step response that looks closer to a theoretical ideal.

The correction from 1) above could be beneficial in case of differing distances between individual speakers and MLP (especially with multi-channel, I expect), but seems that research is fairly sceptical of audible benefits of correction from 2) in realistic (non-anechoic) rooms and with real music signals.
IMHO this is an interesting thread and article on the topic.

Point #2 looks pretty hard to do if I was to just imagine this......makes me think it can't really do point #2 with any accuracy, this is me just shooting from the hip though. I'll let others argue whether it could really do this.

 

[dominikz]

Point #2 looks pretty hard to do if I was to just imagine this......makes me think it can't really do point #2 with any accuracy, this is me just shooting from the hip though. I'll let others argue whether it could really do this.

It can be done with DSP by adding some time delay to the signal and using appropriate non-causal (look-ahead) digital all-pass filters.
EDIT: Maybe this article is helpful if you're interested in some more details on this type of phase correction.

You can even see this works pretty well in Dirac Live from in-room measurements I posted earlier in this thread:

This time I'm showing step response (thanks @mitchco for the suggestion!). We see that Dirac is doing its time-domain magic here, Reference4 exhibits some pre-ringing and the rest don't seem to care much about the time-domain

 

[Robbo99999]

It can be done with DSP by adding some time delay to the signal and using appropriate non-causal (look-ahead) digital all-pass filters.
EDIT: Maybe this article is helpful if you're interested in some more details on this type of phase correction.

You can even see this works pretty well in Dirac Live from in-room measurements I posted earlier in this thread:

Thanks, I'll check that out,.....on a casual level I find it hard to know how software can infer/seperate what's coming from woofer or tweeter just from a frequency sweep, which is (I guess) what the software would have to work out in order to correct anything related to crossover or woofer/tweeter specifics. I'll check your link out soon when I'm in information absorption mode.

 

[Rottmannash]

To my understanding Dirac tries to do two things in the time-domain:
1) Adjust channel delay so that sound from all speakers arrives to the listening spot (MLP) at the same time
2) Correct individual speaker's phase response to compensate for the phase shift introduced by crossover circuit and possible misalignment of drivers. The result is an impulse and step response that looks closer to a theoretical ideal.

The correction from 1) above could be beneficial in case of differing distances between individual speakers and MLP (especially with multi-channel, I expect), but seems that research is fairly sceptical of audible benefits of correction from 2) in realistic (non-anechoic) rooms and with real music signals.
IMHO this is an interesting thread and article on the topic.

Thank you for the explanation. How does Dirac correct based on crossover networks/driver misalignment. Can't visualize how the signal from the amp could be finagled to correct for this.

 

[dominikz]

Thank you for the explanation. How does Dirac correct based on crossover networks/driver misalignment. Can't visualize how the signal from the amp could be finagled to correct for this.

It basically delays different frequencies by a different amount of time. You can see an example in these impulse responses (recorded at DAC output directly):

Here are the impulse responses:

You can see how the LF part of the pulse precedes the HF part in case of Dirac Live.

 

[Rottmannash]

It basically delays different frequencies by a different amount of time. You can see an example in these impulse responses (recorded at DAC output directly):

You can see how the LF part of the pulse precedes the HF part in case of Dirac Live.

Does Dirac need to know how many drivers or the crossover freqs?

 

[dominikz]

Does Dirac need to know how many drivers or the crossover freqs?

The Dirac Live app does not not ask for any details on individual speakers or their layout. I don't really know specifically how their algorithms are designed, but my assumption is they analyze the measured phase response and calculate a correction based on that. Should not be too different in concept vs correction they do to the magnitude (frequency) response IMHO.

 

[Snarfie]

Thanks, I'll check that out,.....on a casual level I find it hard to know how software can infer/seperate what's coming from woofer or tweeter just from a frequency sweep, which is (I guess) what the software would have to work out in order to correct anything related to crossover or woofer/tweeter specifics. I'll check your link out soon when I'm in information absorption mode.

Are we not talkiing about time aligment ?
Above a electronical an physical way to resolve that. So phase shift an time aligment are correlated it looks like. I prefer the physical solution (a more clean signal Path) so i bought Vandersteen speakers. Which requiers lots of moving a round to let them sound as promised. But if they are placed as they shoud be (correct angle an placement from walls) a vail willl be lifted.

 

[thewas]

Are we not talkiing about time aligment ?
Above a electronical an physical way to resolve that. So phase shift an time aligment are correlated it looks like. I prefer the physical solution (a more clean signal Path) so i bought Vandersteen speakers. Which requiers lots of moving a round to let them sound as promised. But if they are placed as they shoud be (correct angle an placement from walls) a vail willl be lifted.

Mind you the chamfer on the baffle adds additional diffraction problems and either you do it electrically or physically the phase alignment only works for the 0° vertical angle unless it is a coaxial driver.

 

[Snarfie]

Mind you the chamfer on the baffle adds additional diffraction problems and either you do it electrically or physically the phase alignment only works for the 0° vertical angle unless it is a coaxial driver.

I'm not a technical specialist but what i understand it takes lots of efforts to get a physically time & phase alignment design working.

This i how Vandersteen (VDS) describes his design :

The Model One uses the proven Vandersteen Aligned Dynamic Design to optimize the dispersion and transient accuracy of the drivers while maintaining the input signal’s time and phase integrity. The drivers, their positioning and their associated minimum baffles were each developed with the aid of FFT (Fast Fourier Transform) computer analysis to minimize diffraction, cone break-up, multi-driver interference and out-of-band phase irregularities.

The construction, alignment and positioning of the drivers allow a point-source wave front and maximize the phase coherence of the loudspeaker at the listening position. The Aligned Dynamic Design is used for the Model One due to its many potential advantages: More precise imaging and a wider listening area. A greater flexibility of placement options within the listening room and better transient response. A high level of genuine transparency and detail typical of planar speakers without the distortions and response variations of multi-directional dynamic loudspeakers. increased efficiency and improved dynamic range. A stable impedance, assuring complete compatibility with any amplifier or receiver.


What i noticed when comparing different same sort off column speakers after correcting them (or the room modes) with Room correction software (Mathaudio Room EQ) is that they all sounded more or less the same regarding stereo image so al instrument are on the same spot but there was still a big difference in staging/depth imaging comparing between the B&W 602 s2, JK acoustic Optima 3 against the VDS model 1. The VDS creates a staging/depth that is far beyond the B&W 602 s2 and/or a JK acoustic Optima 3 despite room correction was used on all this models which should mean that at least phase cohered correction is applied equally for all speakers. Quad ESL electrostatic speakers which i heard where able to create the desired staging but lacked considerably in low frequencies the VDS sounded like an electrostatic speaker (condition is: don't move your head to much put the VDS in the correct angle 1 1/2 inch for around 36 inch of listning height

an placement regarding walls) but could also produce a convincing low.

I don't know if you can say after above comparison that it looks like DSP does not solve all phase en time alignment problems. Basically i don't have the speakers or resources to compare that. But it looks like a hardware solution with the proper order filters can probably do that better than a DSP solution. VDS is using first order filters which is quite nasty to manage an requiers lots of development. The discussions that we have about this subject on ASR comes down that DSP an higher order filters could/would solve the phase coherent / time alignment problems. For now an combination of DSP an VDS hardware solution works fine for me.

On Stereophile there are some interesting interviews with arrived pro musician like Paul Wells, John Escreet, John hHerbert an Billy Drummond. It is interesting that especialy drummers are choosing VDS speakers.
https://www.stereophile.com/content/musicians-audophiles-paul-wells
https://www.stereophile.com/content/musicians-audiophiles-billy-drummond
https://www.stereophile.com/content/musicians-audiophiles-john-escreet
https://www.stereophile.com/content/musicians-audiophiles-john-hébert

 

[hmt]

The problem is not really the delay bn the tweeter and woofer. As the was said that can be easily adjusted but it only affects one angle. Dirac is more about correcting the phase distortion of the crossover. This can be measured and corrected.

 

[phoenixdogfan]

You are right - that would be the best solution. However it is not an option for me at the moment.
As it is, EQ still transforms the sound from unlistenable to quite useable and enjoyable. A lot of the energy in the suck-out area is restored (but not all, of course).
Much better than nothing, in any case!

Sounds to me like you have a room null. AFIK the only way to treat such nulls is multiple subs. A second or even a third sub placed in the "bermuda triangle" area will be the way to treat it.

 

[dominikz]

Several members suggested to have a look at other DRC SW like Audiolense, Acourate and MathAudio, as well as comparisons with manual EQ via REW, e.g.:

Great review, thank you. Any reasons why you did not consider Audiolense or Acourate?

I hope you get a chance to try out Acourate and Audiolense as they are the state of the art DRC with nothing else coming close. Dirac comes a distant 3rd as a "general purpose" DRC as the only thing you can adjust is the target curve (and whether is is full range or partial correction). Dirac uses IIR filters down low and therefore no excess phase correction. Audiolense and Acourate allow full adjustment of all parameters both in the frequency and time domain. With that one can achieve "accurate sound."

Great comparison i was waiting for. Would be interesting to see how Mathaudio Room EQ would compare as a no cost freeware solution.

So I ventured to have a look at all of them, with limited success sadly :

• Audiolense - Went for this one first but didn't get very far Chrome flags the JuiceHiFi website as hosting potentially malicious SW, and even if I go around that, it blocks the Audiolense SW download as potentially malicious. Being my usually cautious self, I decided to wait until they sort all that out before trying to actually install the SW
• Acourate - This is the next one I tried. Not the most user-friendly program to use, but not too difficult either to do basic stuff. Managed to do the in-room measurement and got all the way to successfully creating the inversion vs a target curve as basis for filter generation - alas trial version doesn't let you generate the actual filters from there, so I couldn't test the end-result. This was where I called it quits
• REW + EAPO - Tested this again, and remembered why I gave up last time Anyway got some OK results fairly quickly this time, but only if I avoided EQ boosts. If I used those, listening impressions went south for me quite quickly - and I did try several approaches. Which is quite interesting since I didn't have issues with boosts applied by DRC SW automatically. My takeaway is that either my room EQ skills are still not good enough or my expectation bias is acting up Either way, I'll for sure keep fiddling with it on and off For now I feel there's nothing valuable or new I could report so decided not to do any more writeup on it.
• MathAudio Room EQ - This one I managed to test, some results follow.

MathAudio Room EQ
Tested so far only in my nearfield setup with JBL LSR305s, where I tried a few variants - single-point and multi-point measurements, foobar2000 plugin and running systemwide as a VST plugin in EAPO. First I have to say I loved the usability and simple, easy-to-understand, utilitarian UI.

Measurements and configuration
Running a single-point measurement at MLP I got a similar result as with other tools:

Multi-point measurement across 5 positions around the main listening position was similar:

The tool lets you select two default targets ('Reference'):
"Bright" (flat):

"Neutral" (downward sloping):

But you can also draw your own, which is what I did:

You draw the target curve with the mouse manually in that small window, so you can't really be very precise - but the way it is done is still not too bad and you can do minor corrections to just parts of the curve easily. I tried to follow the natural slope of the speakers, and since this is nearfield, just left a dB or two of low-bass boost.

To my understanding, the SW only corrects the parts of the curve that are above the target, and you should pull it down with the slider on the right until you're happy with the results. I said 'the heck with it' and went all in for my test :

Note that this reduces the overall output by about 12-13dB. The SW allows you to set the bypass signal volume to match this, so you can compare the result with and without the correction with little difference in perceived loudness (depending on how well you tune it).

Listening impressions
First I did some listening and decided I quite enjoyed the initial results. Since MathAudio Room EQ is designed as a full-range correction plugin, it does change the overall sound signature a bit - just like the other DRCs I tried when running them full-range.
Since Dirac Live and ARC3 trials expired for me, I could only compare with Reference 4. So I did a quick compare and I must admit that both Reference 4 and MathAudio sounded comparable to me in this nearfield setup. Not necessarily the same, but both pretty good. Problems area in the bass was much improved in both cases, and the effect on the rest of the spectrum was non-destructive IMHO. The comparison was not blind, nor was the switch between the two easy to do the way I tested, so do take this with a grain of salt.
As mentioned already - you do lose a lot of volume if you run MathAudio with a lot of correction (as I did here) but that is to be expected of any such SW.
Subjectively I was quite impressed - as I expected an over-processed mess considering how much correction I went with - but instead got what I consider fairly natural and enjoyable sound.

Some measurements
Naturally I wanted to see what MathAudio actually did First I ran a loopback test when running MathAudio as VST plugin in EAPO, this was the result (compared to ARC3 and Reference 4):

So MathAudio seems to align closer with ARC3 full-range correction in the LF range, and closer to Reference 4 in the HF range. That tracks with my expectations so all is good as far as I'm concerned
Here's loopback phase response for one channel:

Loopback impulse response:

We see the MathAudio filters are not linear phase and in general I can't see evidence of any phase correction being done here.

Next let's see how it looks in-room at the MLP. First picture shows left speaker, second the right one; below diagrams use variable smoothing to reduce clutter, as most of the interesting details are in the LF range anyway:

That looks scary good to me!

Conclusion
Have to admit I quite enjoyed this first go at MathAudio EQ! It was simple to use and I got what I feel are nice results very quickly. I'm excited to try it out in my living room next to see if I can get comparable results in the far field.

 

[Snarfie]

Several members suggested to have a look at other DRC SW like Audiolense, Acourate and MathAudio, as well as comparisons with manual EQ via REW, e.g.:




So I ventured to have a look at all of them, with limited success sadly :

• Audiolense - Went for this one first but didn't get very far Chrome flags the JuiceHiFi website as hosting potentially malicious SW, and even if I go around that, it blocks the Audiolense SW download as potentially malicious. Being my usually cautious self, I decided to wait until they sort all that out before trying to actually install the SW
• Acourate - This is the next one I tried. Not the most user-friendly program to use, but not too difficult either to do basic stuff. Managed to do the in-room measurement and got all the way to successfully creating the inversion vs a target curve as basis for filter generation - alas trial version doesn't let you generate the actual filters from there, so I couldn't test the end-result. This was where I called it quits
• REW + EAPO - Tested this again, and remembered why I gave up last time Anyway got some OK results fairly quickly this time, but only if I avoided EQ boosts. If I used those, listening impressions went south for me quite quickly - and I did try several approaches. Which is quite interesting since I didn't have issues with boosts applied by DRC SW automatically. My takeaway is that either my room EQ skills are still not good enough or my expectation bias is acting up Either way, I'll for sure keep fiddling with it on and off For now I feel there's nothing valuable or new I could report so decided not to do any more writeup on it.
• MathAudio Room EQ - This one I managed to test, some results follow.

MathAudio Room EQ
Tested so far only in my nearfield setup with JBL LSR305s, where I tried a few variants - single-point and multi-point measurements, foobar2000 plugin and running systemwide as a VST plugin in EAPO. First I have to say I loved the usability and simple, easy-to-understand, utilitarian UI.

Measurements and configuration
Running a single-point measurement at MLP I got a similar result as with other tools:
View attachment 101683
Multi-point measurement across 5 positions around the main listening position was similar:View attachment 101684

The tool lets you select two default targets ('Reference'):
"Bright" (flat):
View attachment 101685
"Neutral" (downward sloping):
View attachment 101686
But you can also draw your own, which is what I did:
View attachment 101687
You draw the target curve with the mouse manually in that small window, so you can't really be very precise - but the way it is done is still not too bad and you can do minor corrections to just parts of the curve easily. I tried to follow the natural slope of the speakers, and since this is nearfield, just left a dB or two of low-bass boost.

To my understanding, the SW only corrects the parts of the curve that are above the target, and you should pull it down with the slider on the right until you're happy with the results. I said 'the heck with it' and went all in for my test :
View attachment 101689
Note that this reduces the overall output by about 12-13dB. The SW allows you to set the bypass signal volume to match this, so you can compare the result with and without the correction with little difference in perceived loudness (depending on how well you tune it).

Listening impressions
First I did some listening and decided I quite enjoyed the initial results. Since MathAudio Room EQ is designed as a full-range correction plugin, it does change the overall sound signature a bit - just like the other DRCs I tried when running them full-range.
Since Dirac Live and ARC3 trials expired for me, I could only compare with Reference 4. So I did a quick compare and I must admit that both Reference 4 and MathAudio sounded comparable to me in this nearfield setup. Not necessarily the same, but both pretty good. Problems area in the bass was much improved in both cases, and the effect on the rest of the spectrum was non-destructive IMHO. The comparison was not blind, nor was the switch between the two easy to do the way I tested, so do take this with a grain of salt.
As mentioned already - you do lose a lot of volume if you run MathAudio with a lot of correction (as I did here) but that is to be expected of any such SW.
Subjectively I was quite impressed - as I expected an over-processed mess considering how much correction I went with - but instead got what I consider fairly natural and enjoyable sound.

Some measurements
Naturally I wanted to see what MathAudio actually did First I ran a loopback test when running MathAudio as VST plugin in EAPO, this was the result (compared to ARC3 and Reference 4):
View attachment 101692
So MathAudio seems to align closer with ARC3 full-range correction in the LF range, and closer to Reference 4 in the HF range. That tracks with my expectations so all is good as far as I'm concerned
Here's loopback phase response for one channel:
View attachment 101695
Loopback impulse response:
View attachment 101696
We see the MathAudio filters are not linear phase and in general I can't see evidence of any phase correction being done here.

Next let's see how it looks in-room at the MLP. First picture shows left speaker, second the right one; below diagrams use variable smoothing to reduce clutter, as most of the interesting details are in the LF range anyway:
View attachment 101693
View attachment 101694
That looks scary good to me!

Conclusion
Have to admit I quite enjoyed this first go at MathAudio EQ! It was simple to use and I got what I feel are nice results very quickly. I'm excited to try it out in my living room next to see if I can get comparable results in the far field.

Hi Dominikz these are really helpful comparisons not only with Mathaudio but in general.
I am using Mathaudio Room EQ now for about 2 ½ year an did tons of multi-point measurements.
I see you use the high resolution measurement mode as i did when it was available as a beta tester December 2018.

Mathaudio Room EQ was the first DSP i used to correct my room reason was that i found REW to complex too use in combination with other software to make it work on a Windows platform.
My experience for now is more than positive first off all about the results the simple way how it operates. For my close monitor setup i use a 9 point measurement in a square meter each point 50 cm. For my full range setup i ended up with a 25 point measurement also in a square meter each point 25 cm. IMsubjectiveO this sounded the best it looks like it revealed more transparency than with lower mulitipoint measurements.

Enclosed some pictures of mine measuring results an preferd target curves which i an other beta testers requested for about 1 ½ ago that Mathaudio added quite soon. The curious thing is when listening to the white flat natural curve i have for my personal taste the best result. Harman curves of B&K curves will cause losses in fine detailed sound like the timbre of voices as especially listening to a full range setup (try Frank Sinatra Live at the Sands an compare his voice with an without target curves quite a difference).
For the close monitor stetting i add some lows (by drawing it) because for my taste it sound better an that drawing feature is really handy.

Would love to hear you findings in your living room positive or negative. I was looking back the last 3 years in my mails an could see that the Mathaudio engine/algorithem is changed/upgraded about 3 times i find his ongoing software change/upgrade quite important for getting a better result.

For Mathaudio Room EQ i changed my amplifier to a NAD C370 which has enough grunth/juice to overcome the low flat target curve. The only disadvantage is dat you realy have watch your volume out put when you change from Foobar2000 to the system wide Windows platfrom. I resolved that to lower my windows out put to a setting of 70% volume at startup. Considering now to buy the paid Mathaudio version for a system wide solution.

Setup:

Vandersteen Model 1C with a flat target curve i prefer.

IMF Compact II with target curve i prefer an draw my self in the low end.

 

[Rottmannash]

Hi Dominikz these are really helpful comparisons not only with Mathaudio but in general.
I am using Mathaudio Room EQ now for about 2 ½ year an did tons of multi-point measurements.
I see you use the high resolution measurement mode as i did when it was available as a beta tester December 2018.

Mathaudio Room EQ was the first DSP i used to correct my room reason was that i found REW to complex too use in combination with other software to make it work on a Windows platform.
My experience for now is more than positive first off all about the results the simple way how it operates. For my close monitor setup i use a 9 point measurement in a square meter each point 50 cm. For my full range setup i ended up with a 25 point measurement also in a square meter each point 25 cm. IMsubjectiveO this sounded the best it looks like it revealed more transparency than with lower mulitipoint measurements.

Enclosed some pictures of mine measuring results an preferd target curves which i an other beta testers requested for about 1 ½ ago that Mathaudio added quite soon. The curious thing is when listening to the white flat natural curve i have for my personal taste the best result. Harman curves of B&K curves will cause losses in fine detailed sound like the timbre of voices as especially listening to a full range setup (try Frank Sinatra Live at the Sands an compare his voice with an without target curves quite a difference).
For the close monitor stetting i add some lows (by drawing it) because for my taste it sound better an that drawing feature is really handy.

Would love to hear you findings in your living room positive or negative. I was looking back the last 3 years in my mails an could see that the Mathaudio engine/algorithem is changed/upgraded about 3 times i find his ongoing software change/upgrade quite important for getting a better result.

For Mathaudio Room EQ i changed my amplifier to a NAD C370 which has enough grunth/juice to overcome the low flat target curve. The only disadvantage is dat you realy have watch your volume out put when you change from Foobar2000 to the system wide Windows platfrom. I resolved that to lower my windows out put to a setting of 70% volume at startup. Considering now to buy the paid Mathaudio version for a system wide solution.

Setup:

Vandersteen Model 1C with a flat target curve i prefer.

IMF Compact II with target curve i prefer an draw my self in the low end.

Why is there a drum levitating in your room?

 

[Snarfie]

Why is there a drum levitating in your room?

Mancave in progress/development

 

[Rottmannash]

Mancave in progress/development

Ah, the old illuminated drum trick. Looks good. Perhaps a Les Paul hanging up there too.