• 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.

Dirac sounds really bad - am I doing something wrong?

tw99

Senior Member
Joined
Mar 13, 2019
Messages
348
Likes
740
Location
West Berkshire, UK
Here is the problem with Dirac, for it to do things that don't completely screw up the natural and proper sound you need a FULLY treated room.
...

In your opinion, clearly.

Personally I like Dirac in my system (untreated room) a lot better than no processing, or using REW derived filters in my MiniDSP SHD.
 

racePigeon

Member
Joined
Jul 17, 2020
Messages
15
Likes
9
Location
Boston area
So did install the Dirac trial today and did the measurement in my finished basement (Highly dumped acoustically).
This the analysis with the default target curve:


default curve.png


I played a few known tracks with and without the filter, and have to say that there's not much difference.
In some cases I actually prefer the sound without the filter.
Have to say that my speakers measure very even, which was not a big surprise.
So I am thinking, I may have saved myself a chunk of money, as I was going to buy a miniDSP device.
I will keep testing in the coming days and probably do other measurements with more positions and see how it goes.
But so far, I think that I don't need to add more processing to the data stream.
 

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
409
Likes
341
Location
Earth
In your opinion, clearly.

Personally I like Dirac in my system (untreated room) a lot better than no processing, or using REW derived filters in my MiniDSP SHD.
sure, and my boss who is a professional acoustician/system designer, and the GLM system from Genelec, and the majority of people I talk to.

Dirac will boost nulls up to 20 db.
This is crazy. I have yet to meet a professional who says that is a good idea.

If your system/room had huge problems with low end then dirac might help, but running it past 400hz will likely ruin the sound.
You may enjoy the ruined sound, but the fact remains its ruined.

Amir likes to EQ out resonances and boost FR to fill in a little missing energy and even with his often conservative and gentle EQ its still hit or miss according to his subjective reviews.

If dirac encounters large variation from your target curve it does horrible things to the sound all in the name of "smooth line"

I bought a miniDSP, i paid a premium for DiracLive, if it did decent things i would say so, but what it does goes against the opinion of every acoustician and system designer i have talked to about it.
 

abdo123

Major Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
4,147
Likes
3,690
Location
Brussels, Belgium
sure, and my boss who is a professional acoustician/system designer, and the GLM system from Genelec, and the majority of people I talk to.

Dirac will boost nulls up to 20 db.
This is crazy. I have yet to meet a professional who says that is a good idea.

If your system/room had huge problems with low end then dirac might help, but running it past 400hz will likely ruin the sound.
You may enjoy the ruined sound, but the fact remains its ruined.

Amir likes to EQ out resonances and boost FR to fill in a little missing energy and even with his often conservative and gentle EQ its still hit or miss according to his subjective reviews.

If dirac encounters large variation from your target curve it does horrible things to the sound all in the name of "smooth line"

I bought a miniDSP, i paid a premium for DiracLive, if it did decent things i would say so, but what it does goes against the opinion of every acoustician and system designer i have talked to about it.

I didn't want to interfere because i don't really care whether someone likes Dirac or not but almost everything you said about Dirac is incorrect.

With Dirac you can build the target curve like a mosaic if that is your wish, and it doesn't boost more than 10 dB. If you want less boost than that you can carve it into the target cuvrve.

What's bad about Dirac is that it 'suggests' measuring 9-11 times instead of measuring each seat once. this way the algorithm actually corrects less at the position that matter because the average amplitude is skewed with areas like the bottom left of your coach.

The other bad thing is that it allows you to correct for frequencies above the transition frequency, unless you have really linear speakers like the LS50 Meta or the Genelec Coaxials you're most likely to ruin the sound that way.

Both of these negative qualities are 'available' but are not really forced upon you. this way you can use what you can from it (the phase correction and room modes correction) and leave
 

Tangband

Addicted to Fun and Learning
Joined
Sep 3, 2019
Messages
595
Likes
619
One can get much better sound if going active, instead of combining dsp eq with passive crossovers.
 

mdsimon2

Addicted to Fun and Learning
Joined
Oct 20, 2020
Messages
594
Likes
642
Location
Detroit, MI
I mean YOU drew the target curve to boost that null at 90 Hz, if you don’t want that boost then change the curve or limit the frequency range of correction. I also see no evidence of Dirac boosting more than 10 dB.

I do not use Dirac personally but not sure your criticisms make much sense.

Michael
 

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
409
Likes
341
Location
Earth
I mean YOU drew the target curve to boost that null at 90 Hz, if you don’t want that boost then change the curve or limit the frequency range of correction. I also see no evidence of Dirac boosting more than 10 dB.

I do not use Dirac personally but not sure your criticisms make much sense.

Michael
Incorrect, Dirac drew the curve, Dirac can and does often limit what it thinks it can do and it will set the lower limit slider based on it's design/algorithm
As I said in post, I let Dirac do it's thing and did not draw or change anything, then I quickly created a filter set to mic it.

Regardless of what decisions for under 400hz Dirac makes, the Filter design shows the issues above 400hz.
 

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
409
Likes
341
Location
Earth
I didn't want to interfere because i don't really care whether someone likes Dirac or not but almost everything you said about Dirac is incorrect.

With Dirac you can build the target curve like a mosaic if that is your wish, and it doesn't boost more than 10 dB. If you want less boost than that you can carve it into the target cuvrve.

What's bad about Dirac is that it 'suggests' measuring 9-11 times instead of measuring each seat once. this way the algorithm actually corrects less at the position that matter because the average amplitude is skewed with areas like the bottom left of your coach.

The other bad thing is that it allows you to correct for frequencies above the transition frequency, unless you have really linear speakers like the LS50 Meta or the Genelec Coaxials you're most likely to ruin the sound that way.

Both of these negative qualities are 'available' but are not really forced upon you. this way you can use what you can from it (the phase correction and room modes correction) and leave
Is there a way in Dirac LIVE to get more than the low/high limit sliders and make a mosaic?
That may be a feature on the software but I have looked and not found it on hardware linked version.
 

mdsimon2

Addicted to Fun and Learning
Joined
Oct 20, 2020
Messages
594
Likes
642
Location
Detroit, MI
It drew it as a default which you are free to change. Just find it funny you are complaining about boosts when you could have prevented this by modifying the curve to something that you think is more appropriate.

You claimed 20 dB of boost, all information to date from miniDSP has indicated that boosts are limited to 10 dB (this is also what your measurements show). Still looking for evidence of this supposed 20 dB of boost.

Again, l don’t use Dirac but this seems more like the general learning curve of the tool than any inherent flaws.

Michael
 

abdo123

Major Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
4,147
Likes
3,690
Location
Brussels, Belgium
I don't have time to get into all the details today, but for sake of simplicity, I wanted to show an "average use" case.

Filter set 1.
Dirac 1 position measurement at desktop in my bedroom
Measured at 90dbc to allow 40db over noise floor.
Gain adjust on mic to have 18db headroom.
no clipping found

For this average test case I left the target slope as Dirac and changed nothing.
Here is the acoustic Dirac measurement AND the pure miniDSP output to verify electronic filter assignment.


View attachment 156418


And here is the electronic measurement straight from miniDSP lookback into zoom h6

View attachment 156412

Now, I dislike everything Dirac does BUT in order to compare apples to apples, I quickly recreated the Dirac target curve
And set REW filter design to "max boost" on everything and "1db to target" to duplicate, as close as possible what Dirac does.
I must repeat, I would NEVER set any system up this way.
Here is the filter design and electronic loopback confirmation

View attachment 156413

Loopback


View attachment 156415

So to get REW to do radical things like Dirac always does when running full range I had to set REW to "max all" and "1db to target"

To compare the results side by side I lowered one and raised another to see the different, Here it is

View attachment 156416

The jagged, messed up and non-musical Filter set up top is Dirac.
The smooth, musical, but still not recommended set on bottom is REW

You can argue all you want that the Dirac set makes sense, in my opinion and those of other professionals it doesn't. What Dirac results in is a HUGE loss of headroom and mega distortion when it fills in LF nulls and also unnatural voices trumpets, and cymbals, basically everything above 300hz when it applies narrow boosts that are very audible.

But if you like the top, Dirac filter design than good, after all, it's all about liking what you pay for!

Really? for me REW looks like is boosting a shit ton more than Dirac. REW is also doing high Q filters very high in the Frequency range While Dirac is not.

REW is limited with the 10-20 PEQ filters that it is assigned, if you increase the limit so the resolution of Dirac's mixed phase approach would match REW's resolution then you will probably have very similar outputs.

Dirac does not do anything 'special' to the magnitude response other than correcting the average of all the measurements you took to the target curve you set. it's not worse or better than REW in that regard except that it doesn't give you the option for 'max boost' like REW does.

It seems like you were expecting some Voodoo magic and instead you're surprised that this technology does not break the laws of physics and will eat up headroom, like every EQ out there, and will sound wrong if you used it incorrectly.

From my perspective the 'noob' first timer (well that's what i did at the begining) will indeed take all the measurment points suggested by the program. The resulting average will very much resemble the Predicted in-room response of the speakaer above 300Hz or so. So whatever corrections that happen are rarely actually high Q unless the speaker has a high Q resonance.

Below 300Hz the response will be indicative of the room and boundary interference, but since so many measurements were taken barely any high Q modes will be shown other than the axial room modes and the boundary interference.

Thus the 'noob' experience will be positive but not very great. unless they eat all the headroom they have boosting below 300Hz then it will sound off.

However the 'expert' (take only as much measurement points as you actually need and correct only below transition frequency while paying attention to the amount of boost you're applying) experience is very positive relative to the time you spent setting it up.
 

abdo123

Major Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
4,147
Likes
3,690
Location
Brussels, Belgium
Is there a way in Dirac LIVE to get more than the low/high limit sliders and make a mosaic?
That may be a feature on the software but I have looked and not found it on hardware linked version.

you add a point to the target curve by right clicking (or left clicking) in the middle of the curve on the frequency you want adjusted.
 

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
409
Likes
341
Location
Earth
It drew it as a default which you are free to change. Just find it funny you are complaining about boosts when you could have prevented this by modifying the curve to something that you think is more appropriate.

You claimed 20 dB of boost, all information to date from miniDSP has indicated that boosts are limited to 10 dB (this is also what your measurements show). Still looking for evidence of this supposed 20 dB of boost.

Again, l don’t use Dirac but this seems more like the general learning curve of the tool than any inherent flaws.

Michae

It drew it as a default which you are free to change. Just find it funny you are complaining about boosts when you could have prevented this by modifying the curve to something that you think is more appropriate.

You claimed 20 dB of boost, all information to date from miniDSP has indicated that boosts are limited to 10 dB (this is also what your measurements show). Still looking for evidence of this supposed 20 dB of boost.

Again, l don’t use Dirac but this seems more like the general learning curve of the tool than any inherent flaws.

Michael
This is simple test case to observe the difference in filter set design.
Regardless, if Dirac only has a low and high limit bar you can find yourself in this position.

1. Need to lower a peak at 40hz
2. Large null at 80hz

In order for Dirac, with only 2 sliders to adjust, If you set the slider low enough for the 40hz peak it will automatedly try and raise the null.

Multiple "limit sliders" would fix this issue but the fact remains Dirac uses narrow filters to boost, and does so in a way that is against standard EQ filter set design in the music production industry.
 

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
409
Likes
341
Location
Earth
you add a point to the target curve by right clicking (or left clicking) in the middle of the curve on the frequency you want adjusted.
I've tried that many times, to add points to fallow the natural frequency response.
I'll to that again when I get back and see how little Dirac will affect it in the places it shouldn't.
 

abdo123

Major Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
4,147
Likes
3,690
Location
Brussels, Belgium
Multiple "limit sliders" would fix this issue but the fact remains Dirac uses narrow filters to boost, and does so in a way that is against standard EQ filter set design in the music production industry.

I really think you should read a little more about the service, the filters and the way they're implemented are not standard in anyway, it's basically FIR correction on a budget. the filters are designed in a way that they can run on very low processing power devices as long as the computer generating the filters is actually powerful. This way the 4000 taps FIR filtering (linear phase) and 10-PEQ filters (minimum phase) in the MiniDSP 2*4 are 'combined' to generate a Dirac filter (mixed phase) with 48000 taps resolution. Which is the amount of resolution any person would realistically need for correction.

The magnitude response correction is in-fact 'optional' in a sense, they implemented it because they knew people would want it not because they think it's absolutely necessary. At least that's the vibe i caught When Erin interviewed a big shot at Dirac.

 

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
409
Likes
341
Location
Earth
I really think you should read a little more about the service, the filters and the way they're implemented are not standard in anyway, it's basically FIR correction on a budget. the filters are designed in a way that they can run on very low processing power devices as long as the computer generating the filters is actually powerful. This way the 4000 taps FIR filtering (linear phase) and 10-PEQ filters (minimum phase) in the MiniDSP 2*4 are 'combined' to generate a Dirac filter (mixed phase) with 48000 taps resolution. Which is the amount of resolution any person would realistically need for correction.

The magnitude response correction is in-fact 'optional' in a sense, they implemented it because they knew people would want it not because they think it's absolutely necessary. At least that's the vibe i caught When Erin interviewed a big shot at Dirac.

Thanks for the vid.
I knew they had talked about using FIR filters to adjust time domain in regions but I had never heard them explain they were also adjusting globally on both sides of the crossover to make the woofer and tweeter match on impulse response better, I do not believe this can be done in a musical and good sounding manner the way the Dirac creator, Jakob, described.

It's one thing to use DSP/amps and to time align the tweeter/woofer separately but Jakob said his algorithm is doing this globally. He says if the woofer is slow he is delaying the tweeter so that the woofer and tweeter will have a cleaner impulse response and arrive together. There is only so much you can delay each side leading into the crossover region before any delay would equally affect both drivers.
Wouldn't any delay at a 2500hz xover overlap point , if that was the frequency, equally affect the sound of both driver, but a delay at 3000hz will affect the tweeter more, even though a little woofer response may still be relevant.

BBE sonic maximizers claim "This function is used to "time-align" the different transducers in a speaker system. A properly aligned system will have better phase coherence resulting in more focused sound"

Problem is, with speaker design, BBE can't know which driver's sound is arriving first, although BBE are used often in studios on kick and snare, they aren't used for much else.

Since this is what Jakob is saying Dirac is doing, it brings up some interesting questions.

1. How does this time delay work inside and outside of the crossover region.
2. How does the delay work on 3 way towers.
3. How does the crossover slopes of woofers and tweeters affect this process.

It wouldn't surprise me that these things have a big effect on how pleasant or musical the result is.
As it stands now I do not like what Dirac does, I do not like the filter design, the choices it makes, or its customization, lack of.
It's filter design is fixed. If you want it to process any region it will do so for magnitude and time domain, both of close sections and of tweeter to woofer.

Just as I don't think BBE's singular approach can "correct the time arrival of different transducers" I also don't believe what Dirac and Jakob are doing can be done musically. Dirac CAN get the time alignment closer between transducers but at a cost I believe to be the loss of the natural sound of instruments and musicality of the sound system.

Thanks for the Vid ABDO123.

And on recheck the highest boost Dirac did was 10-12db, so 20db was a loose exaggeration.
 
Last edited:

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
409
Likes
341
Location
Earth
Looking to see what others said about Dirac's time alignment of woofer/tweeter others surmised the same thing.
I know nothing about this other forum and am not claiming the statement is anything but other's surmising.
But that's the issue with hidden/propriety filter design, all we can do is surmise.


"My guess is:

Dirac can only "fix" time alignment when it has complete control of a particular driver, e.g. in the middle of a driver's passband. Then it can advance or delay the signal. In the crossover frequencies, it can not because it is the xo circuits that determine that.

However, it can still use it's phase and amplitude fixing powers to mostly patch it up.

I think the general advice is to try to do the time alignment first outside of Dirac."

As far as I know, and as others have implied, there is no way to time align a woofer/tweeter the way Dirac does without without causing time domain distortions when sound passes from one side of crossover region to the others.
Perhaps this is why cymbals sound so bad to me with Dirac?
If sounds are fully in tweeter, fully in woofer, or full/exactly in xover region then the Dirac method could work, but that isn't how music is. In music sounds travel the whole spectrum. Delaying part of that sound relative to Xover region is going to cause issues. Issues I say are some of the worst encountered with Dirac.

This makes perfect sense as why some love Dirac and others don't.
I believe desktop use of taller bookshelf speakers and also designs with very reassessed or forward physical speaker alignment would force Dirac to apply more delay to "fix" it and thus causing more Xover/time domain issues.

A simple addition of
"do not time align drivers"
and
"Flatness target curve"
Would allow Dirac users to have options and to see what separate objectives of the algorithms sound best.
 
Last edited:

abdo123

Major Contributor
Forum Donor
Joined
Nov 15, 2020
Messages
4,147
Likes
3,690
Location
Brussels, Belgium
Dirac CAN get the time alignment closer between transducers but at a cost I believe to be the loss of the natural sound of instruments and musicality of the sound system.

I'm a bit busy today so i can't reply to all the points you mentioned right now but i suggest you give Dirac a sensible try first (limit magnitude correction to transition frequency, only measure what needs to be measured, adjust target curve if you run out of headroom) before we continue this discussion.
 

raindance

Addicted to Fun and Learning
Joined
Sep 25, 2019
Messages
549
Likes
492
It would be interesting to see a picture of your room layout. You mention a bass trap behind a speaker. This could mean your speakers are in corners because that's where bass traps are supposed to be. Corners will give massive bass bloat and bass traps won't fix this, they'll just change the decay time at the frequencies they work at, which depends on how large they are and what they're designed to do. The typical fairly thick wall mounted panels called bass traps aren't very effective at low frequencies, only really upper bass. To get down to the 40Hz range you need some pretty big corner bass traps, floor to ceiling, and even then, they'll just shorten decay times more than change actual frequency response.

In my room, I found adding 2 of the largest GIK free standing bass traps in each front corner made the mid and upper bass sound cleaner, shortened the decay of the primary modes in room, and made very small, almost negligible changes to measured frequency response in the bass area. So the outcome wasn't to eliminate the room modes, it was to lessen their effect on other frequencies.

Someone asked earlier if it was possible that DIRAC is ignoring room treatment. Well no, this isn't possible, it only "sees" what the mic sees. It isn't sentient.

Turning up the subwoofer after correction kind of defeats the object of having a preference curve and isn't a solution in my book, FYI.
 
Top Bottom