Dirac Live Exclamation Mark issue

[Sam1879]

Sorry, i didn‘t read everything exactly, just wanna report my findings: Same Problem one month ago on DDRC22d with MACBook and UMIK2 after Dirac (and maybe MACOS) Update. After some trys without sucess I tried on an older Windows Notebook with newest DIRAC and Win 10. Worked perfectly, sounds wonderful now.

 

[Davide]

I have been running Dirac for years via a plug-in (using Jriver) and never encountered a thump such as you describe.

I'm sure mine is an unfortunate case.
I tried with two different DAWs, Reaper (which does the thump) and now Element, but here too there is a very loud artifact before the sweep...
It seems a buffer flushing problem...

 

[Davide]

Sorry, i didn‘t read everything exactly, just wanna report my findings: Same Problem one month ago on DDRC22d with MACBook and UMIK2 after Dirac (and maybe MACOS) Update. After some trys without sucess I tried on an older Windows Notebook with newest DIRAC and Win 10. Worked perfectly, sounds wonderful now.

What problem do you encounter exactly?

 

[Sam1879]

What problem do you encounter exactly?

I wanted to remeasure my room with the newest version of Dirac Live, updated it to V 3.9.7 (same Problem after downgrade to V3.8.2) on my MacBook Pro (15-inch, 2016, MacOs 12.7.x) and had the following problem:
When in the step of calibrating the volume of the speakers before doing the measurements, the bar showed the volume level was way too high at volume levels that were good / in the "green" region at earlier versions.
When I then started a measurement to measure with lower or especially my normal measurement-volumes, the graphical animation which shows the frequency/volume "curve" during the measurement swipes was looking like it was 80-100% at full volume on the whole spectrum. Like much too loud or totally compressed.
If i remember right, most or all measurements showed the exclamation mark, there was a yellow notice that the measurement was bad. Too much distortion.
I don't remember if i could do one valid measurement.
Reinstall didn't help. If I remember right, there was the same problem with UMIK1, but it would have worked correctly with my Macs internal microphone.

Everything worked fine on the first atempt with the windows laptop.

 

[Davide]

From what I have read online the umik1 often introduces clock problems. There was a thread on some forums that showed extreme inconsistency and variability between different units.
Plus it works on usb interface, so clock drift may occur with respect to playback device.
In addition, his ADC works at 48kHz fixed, so working with different sample rate implies resampling somewhere (that if it is not linear phase affects the IR).

You who have the problem of the exclamation point, have you tried to investigate in this sense?

In the meantime, I also discovered that if I do the measurements with the Dirac plugin there is an absurd rolloff at high frequencies...

 

[Esprit]

No response from Dirac's team?

 

[bungle]

No response from Dirac's team?

I feel Dirac team is not in control of their software, that it is out sourced, or that they do no have competent programmers. Their answers almost always sound like ”they need to ask from someone else” (it's a broken phone).

 

[Great Sun Jester]

I feel Dirac team is not in control of their software, that it is out sourced, or that they do no have competent programmers. Their answers almost always sound like ”they need to ask from someone else” (it's a broken phone).

I wonder if they are caught between a rock and a hard place. They want ART functionality on everything but don't want to tell some, tough luck. If you can load the new software than you can probably run ART but what I don't understand is why some can revert back and it still doesn't work.

 

[Esprit]

I feel Dirac team is not in control of their software, that it is out sourced, or that they do no have competent programmers. Their answers almost always sound like ”they need to ask from someone else” (it's a broken phone).

I bought a Dirac (haven't tried it in months yet), I have an Umik, I have a Mac and reading this thread makes me want to sell it.
PS I tried Dirac many years ago and it worked fine on the Mac. Reading all your problems depresses me.

 

[Davide]

Let's say that currently 729/849 € for the Studio version with Bass Control is a bit of a theft if you find yourself with an ineffective correction and a technical support that is hiding in finding valid solutions or explanations.
Some personal experiences indicate this, but there are also good ones.
You should try before selling.

 

[Esprit]

As I wrote, I used it a lifetime ago when they were "unknown".
I purchased the Stereo version on promotion in November but for a thousand reasons I never had time to use it.
Reading the above doesn't make me happy anyway...

 

[RenPa]

I feel Dirac team is not in control of their software,

I feel the same. They should have a certification procedure to control the implementation of ther software in a brand of reciver

They want ART functionality on everything but don't want to tell some, tough luck.

I would bet on that. Since ART capabilities have been introduced in 3.5.3 update firmware, DLBC optimization calculated filters can't be trusted on Arcam's and some others brands Harman related. We are waiting a fix since near a year now...

Reading the above doesn't make me happy anyway...

But to be fair, when it works it is worth it.

 

[Davide]

I did a test to verify the behavior of the Dirac algorithm and a strange thing emerged (my configuration is 2.2).

I took my current target and created a filter with DLBC.

Then I took the same target and using Excel I shifted it downwards uniformly by 6.5dB, so that the highest point (i.e. the boost at the low end) corresponded to 0dB of target.

Next I captured the processor output with digital loopback from Blackhole and REW and these are the GD results.
FILTER2 is the one attenuated by 6.5dB (yellow/orange traces).

Three things can be noted:
- The first is that there is a reduction in the GD introduced by the filter at the low end.
- The other is that the alignment of the subs compared to the mains completely changes. With the attenuated target the subs are brought forward by 20ms, while the mains remain almost the same.
- The GD difference between the 30Hz range and the 100Hz range increased with the new filter (and seeing the acoustic response of the subs, this does not correspond to an improvement in the correction but to a worsening).

There are two explanations in my opinion:
- Dirac works in mixed phase so the difference that appears serves to obtain alignment with the phases resulting from the different magnitude correction.
- Dirac gets the alignment badly wrong.

However, the point of the test is that, if the attenuated filter provides better GD due to mixed phase operation, it is sufficient to apply 6.5dB gain elsewhere, even trivially with the volume knob, to obtain the exact same SPL as the first filter.

As soon as I can I will carry out acoustic measurements to see the actual result of these filters.

EDIT: Acoustic measurements confirm the behaviour of digital loopback results.

EDIT2: Results of other investigations.

 

[Rottmannash]

Sorry, i didn‘t read everything exactly, just wanna report my findings: Same Problem one month ago on DDRC22d with MACBook and UMIK2 after Dirac (and maybe MACOS) Update. After some trys without sucess I tried on an older Windows Notebook with newest DIRAC and Win 10. Worked perfectly, sounds wonderful now.

I'm not sure if it's germane but I'm using an older HP laptop running Windows 10. Like many, I had no problems running the OG 3.4.4 but none of the newer versions work without endless fiddling with levels and re-running sweeps. Even now when reverting to 3.4.4 the results are very different with the same mic placements and same # of positions. Can't imagine why using Win11 vs 10 would make a difference but who knows? (Dirac?)

 

[mstanley117]

I'm not sure if it's germane but I'm using an older HP laptop running Windows 10. Like many, I had no problems running the OG 3.4.4 but none of the newer versions work without endless fiddling with levels and re-running sweeps. Even now when reverting to 3.4.4 the results are very different with the same mic placements and same # of positions. Can't imagine why using Win11 vs 10 would make a difference but who knows? (Dirac?)

My best, educated guess regarding the Dirac imprecise measurement issue, and this is entirely based on my anecdotal experience, is that the current versions of Dirac are computer hardware dependent, Basically, I need to run fairly modern computer hardware (latest mobile Ryzen in my use case) to avoid imprecise measurement issues.

To add further context: all of the computers I have use Windows 10 but LTSC 21H2, so not the most up to date W10. I don't have any Apple products, and I'm currently not yet using Windows 11 (until LTSC comes out but that's not relevant to this discussion). All of the computers I've used are mobile-based (my mini-desktop, a 2019 HP Envy x360, and a 2024 Lenovo Legion - this is the only one that gets successful Dirac results, btw).

This is also in reference to only Dirac 3.9.7. I use a Umik-1 and a MiniDSP-SHD for Dirac, and I cannot use a USB extension cable at all for the Umik-1. I used to be able to use the extension cable, and Dirac used to be able to return positive results on all of my computers prior to 3.9.7. I have not tried rolling back to Dirac 3.4.4.

Anyway, I couldn't tell you if it's a USB Mic issue, a clock issue, etc. The only reasonable conclusion I've drawn for myself is that I have a moderately powerful laptop that I have to use to run Dirac with my hardware to get any beneficial results.

 

[RenPa]

I did a test to verify the behavior of the Dirac algorithm and a strange thing emerged (my configuration is 2.2).

I took my current target and created a filter with DLBC.

View attachment 368767

Then I took the same target and using Excel I shifted it downwards uniformly by 6.5dB, so that the highest point (i.e. the boost at the low end) corresponded to 0dB of target.

View attachment 368768

Next I captured the processor output with digital loopback from Blackhole and REW and these are the GD results.
FILTER2 is the one attenuated by 6.5dB (yellow/orange traces).

View attachment 368769

Three things can be noted:
- The first is that there is a reduction in the GD introduced by the filter at the low end.
- The other is that the alignment of the subs compared to the mains completely changes. With the attenuated target the subs are brought forward by 20ms, while the mains remain almost the same.
- The GD difference between the 30Hz range and the 100Hz range increased with the new filter (and seeing the acoustic response of the subs, this does not correspond to an improvement in the correction but to a worsening).

There are two explanations in my opinion:
- Dirac works in mixed phase so the difference that appears serves to obtain alignment with the phases resulting from the different magnitude correction.
- Dirac gets the alignment badly wrong.

However, the point of the test is that, if the attenuated filter provides better GD due to mixed phase operation, it is sufficient to apply 6.5dB gain elsewhere, even trivially with the volume knob, to obtain the exact same SPL as the first filter.

As soon as I can I will carry out acoustic measurements to see the actual result of these filters.

EDIT: Acoustic measurements confirm the behaviour of digital loopback results.

EDIT2: Results of other investigations.

Soory I struggle to see. Too many plots in your image to clearly see what you mean. Do you suggest to always use a shifted downward house curve? BTW your findings seem to fit with others in other forum about erratic DLBC behavior with group delay. For one, I found that the only usable DLBC version for my receiver is 3.4.4 .

 

[Davide]

Soory I struggle to see. Too many plots in your image to clearly see what you mean. Do you suggest to always use a shifted downward house curve? BTW your findings seem to fit with others in other forum about erratic DLBC behavior with group delay. For one, I found that the only usable DLBC version for my receiver is 3.4.4 .

The point is that Dirac is inconsistent in generating filters.
Furthermore, I found that for each specific combination of target, crossover, curtain, etc, there are two different filters that it can calculates, dependent only on the iteration.
That is to say: you calculate a filter, export it, then move for example the crossover manually and then put it back in the exact same position as before, recalculate, export, and you will see that by doing REW digital loopback measurements there are different behaviour, especially in time domain.
The same phenomenon is sometimes evident also from the Dirac graph where it shows the estimate of the correct response. In one interaction you see some dips and not in the other.

I reported this to technical support and in addition to telling me that I was not the only one to report it, they told me that it is a sort of natural behavior of the algorithm.
I can no longer recover the exact answer because they changed the technical support platform.
But was it something like the example of circles that propagate in the water when you throw a stone...

PS. Doing a thousand experiments and selecting the best filter, that is one with less group delay, the result is exceptional. I really like how it sounds.
With Audiolense I can't even get close to the consistency of the Dirac' bass and IACC value.

The fact of not having a perfectly linear phase system, including crossovers, is acoustically negligible to my ears.
The only thing that I hear is delayed low end, that seems like blurred bass, not tight.

I don't want to get the message across that it's bad software. Apart from the aforementioned problems it works exceptionally, especially considering the practicality.

 

[ppataki]

The point is that Dirac is inconsistent in generating filters.
Furthermore, I found that for each specific combination of target, crossover, curtain, etc, there are two different filters that it can calculates, dependent only on the iteration.
That is to say: you calculate a filter, export it, then move for example the crossover manually and then put it back in the exact same position as before, recalculate, export, and you will see that by doing REW digital loopback measurements there are different behaviour, especially in time domain.
The same phenomenon is sometimes evident also from the Dirac graph where it shows the estimate of the correct response. In one interaction you see some dips and not in the other.

I reported this to technical support and in addition to telling me that I was not the only one to report it, they told me that it is a sort of natural behavior of the algorithm.
I can no longer recover the exact answer because they changed the technical support platform.
But was it something like the example of circles that propagate in the water when you throw a stone...

PS. Doing a thousand experiments and selecting the best filter, that is one with less group delay, the result is exceptional. I really like how it sounds.
With Audiolense I can't even get close to the consistency of the Dirac' bass and IACC value.

The fact of not having a perfectly linear phase system, including crossovers, is acoustically negligible to my ears.
The only thing that I hear is delayed low end, that seems like blurred bass, not tight.

I don't want to get the message across that it's bad software. Apart from the aforementioned problems it works exceptionally, especially considering the practicality.

It is great indeed if/when it works
The problem is - and that's why I started this whole thread - that it does not work properly (for some users, including me)

In my opinion this is not acceptable (especially given the fact that we are talking about a payable software...) and should have been fixed already like 6 months ago

 

[RenPa]

Doing a thousand experiments and selecting the best filter, that is one with less group delay, the result is exceptional. I really like how it sounds.

A question. Although I know REW and understand the basics of Group Delay , if I want to check a DLBC filter GD with REW do I measure Front left only with the filter on so I measure the redirected bass in the subs and the rest of the SPL in the speaker all together or do I disconnect the Front speaker, measure the subs, reconnect the speaker and remeasure?

 

[Davide]

A question. Although I know REW and understand the basics of Group Delay , if I want to check a DLBC filter GD with REW do I measure Front left only with the filter on so I measure the redirected bass in the subs and the rest of the SPL in the speaker all together or do I disconnect the Front speaker, measure the subs, reconnect the speaker and remeasure?

For DLBC, you just need to measure a single channel (example L), without disconnecting anything.
To judge the behavior of the filter, however, you must also measure the sub(s) without Dirac, so you must be able to make the measurement by sending the signal from REW directly to the sub(s).
To compare the measurements correctly you must use the same acoustic reference.

What configuration do you have?