• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Dirac Live 3 vs. Manual Effort with my MCLA

OP
P

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,216
Likes
1,356
Location
Budapest
@ernestcarl
Just one question in the meantime: I see your settings in RePhase:
1651849124307.png


Are these the 'optimal' settings? Shall I also keep sticking to them?

What I am doing is that I am importing the REW filters into RePhase using the 'import REW filters settings' option

I see the dotted blue line representing the phase if I am not mistaken:

1651849305217.png


Now if I start fiddling around with the Paragraphic Phase EQ to try to 'linearize' the phase curve I guess that would be beneficial, right?

Something like this:
1651849637506.png


Can you please comment on this?
Many thanks
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
You can only use settings like that if you are using a pure minimum phase filter. If it is minimum phase you can also put zero in the centering box and turn off optimization for the lowest latency. Anything from 16384 to 65536 has enough resolution particularly at 48K.

If you try and alter the phase response this goes out the window and the impulse will have to be much closer to the centre of the filter and that will introduce latency. It is the shifting of the impulse centre and latency that allows the "time travel" to go back and alter things before they have happened :)

Arbitrarily trying to flatten the phase is not a good strategy. What you want to aim for is a response that gets as close to minimum phase for the magnitude response as you can. It's important to know where and why the response is becoming minimum phase. You can extract excess phase from REW import that and try and flatten it, but trying to fix things that should be left alone or need an acoustic solution may not improve the sound.
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Are these the 'optimal' settings? Shall I also keep sticking to them?

Well, it should work... you can use less taps and change some of the parameters like the sampling if you actually use 44.1k or 96k etc. The red and blue lines will tell you how well the generated outcome matches the desired prediction respectively. I added -20dB global negative gain in the first tab (a little above the maximum dB gain/boost used) which you can change if you want.

What I am doing is that I am importing the REW filters into RePhase using the 'import REW filters settings' option

No reason to re-import that into rePhase itself as the filters are already saved in the left and right rePhase config files.

Fluid is right in that minimum phase EQ filters is all that is really needed... so essentially no additional processing delay required.

You can try some linear phase EQ filters (it's not needed), and you will likely incur ringing. It's pretty easy to test out by switching the paragraphic gain settings per bank to "linear phase" and change the centering to "middle". Run a sweep with the generated filters and see for yourself.

There is very little excess phase to flatten out in your equalized measurements in the bass region to begin with... but it can be done -- not worth it, IMO, but still doable.

If you want to flatten whatever leftover excess phase is there in the bass...

"Excess phase version" is generated in REW and exported as text and imported or easily dragged (file itself) directly into rePhase. In the All SPL options tab... click the 'measurement actions' button. Mind you, some windowing and/or smoothing may need to be applied to make the phase workable inside rePhase.
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
When I started to try and make the filters I realised I need your mic calibration file to use in DRC. If you can post the txt file I can make them.
 
OP
P

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,216
Likes
1,356
Location
Budapest
When I started to try and make the filters I realised I need your mic calibration file to use in DRC. If you can post the txt file I can make them.
Please see attached
Thank you
 

Attachments

  • 7013935_90deg.zip
    4.7 KB · Views: 39
OP
P

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,216
Likes
1,356
Location
Budapest
You can only use settings like that if you are using a pure minimum phase filter. If it is minimum phase you can also put zero in the centering box and turn off optimization for the lowest latency. Anything from 16384 to 65536 has enough resolution particularly at 48K.

If you try and alter the phase response this goes out the window and the impulse will have to be much closer to the centre of the filter and that will introduce latency. It is the shifting of the impulse centre and latency that allows the "time travel" to go back and alter things before they have happened :)

Arbitrarily trying to flatten the phase is not a good strategy. What you want to aim for is a response that gets as close to minimum phase for the magnitude response as you can. It's important to know where and why the response is becoming minimum phase. You can extract excess phase from REW import that and try and flatten it, but trying to fix things that should be left alone or need an acoustic solution may not improve the sound.

Can you guys please shed some more light on the centering and on the optimization options?

What is the impact of putting there 0ms or any other values?
I am trying to understand this a bit more
Many thanks
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Can you guys please shed some more light on the centering and on the optimization options?

What is the impact of putting there 0ms or any other values?
I am trying to understand this a bit more
Many thanks

"0" will not work with the last two HF PEQs used -- bypass the two and it will work.

Honestly, don't know enough of the details of it myself... but the ff. images may shed some light why "0" doesn't work with the two HF PEQs:

1651937795841.png


There is still some activity of the filter prior the apparent 0 marked beginning of the impulse -- not peak itself.

0.1ms seems to work well enough -- but there is still some loss of accuracy in the magnitude response.

1651937876232.png

17 & 19kHz boosting PEQs disabled


----



Linear phase HP and LP for subwoofer:

1651938624427.png


Delay is 170.67ms still with some inaccuracy in the magnitude and phase.

If we look at the filter's IR & step:

1651938689629.png


You can roughly intuit maybe why that is the case...
 
Last edited:
OP
P

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,216
Likes
1,356
Location
Budapest
0.1ms seems to work well enough -- but there is still some loss of accuracy in the magnitude response.

At least to me the 0.1ms IR (red) looks much better than the yellow above it since its post-ringing is way smaller
Am I correct with that statement?
If yes then I guess this might be even an audibe difference - I will have to test it
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
At least to me the 0.1ms IR (red) looks much better than the yellow above it since its post-ringing is way smaller
Am I correct with that statement?
If yes then I guess this might be even an audibe difference - I will have to test it

Nah... it's too low in level to be audible, I think.

Just from a visual inspection of the wavelet:

wave ripple 1.png wave ripple 2.png
*yours (yellow legend) is a convolved prediction with the EQ filters

You could always zoom-in to any of these graphs with a fine microscope -- but, it doesn't mean human ears perceive these fluctuations in the graphs like a computer/machine can.


Last time I heard ringing with very careful listening had a wavelet that looked more like this:

1651950287893.png
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
Can you guys please shed some more light on the centering and on the optimization options?

What is the impact of putting there 0ms or any other values?
I am trying to understand this a bit more
Many thanks
Just to be clear it is 0 (sample value) and not 0ms. By using 0 samples and closest perfect impulse and optimization to none you should get a perfect impulse as close to 0 samples as possible. In this the example case below it was 0.003 samples. There is no need for optimization unless you are trying to use a smaller number of taps than the filter really needs. In ernestcarl's example above with a high pass 16384 is not enough taps to match the desired slope, but as it only diverges slightly under 10Hz the reduction in latency could be a good trade off.

Optimization sounds like it will always give an improvement which is not necessarily so. I would turn it off unless you are trying to reduce the tap count. The window used will have a big impact on how optimization works, some windows work well when trying to really squeeze the tap count others very badly.

Centering to a specific time can be helpful when using filters for multiways where you want to keep the latency the same for all ways and not not have it vary in each filter.

It doesn't make sense to look at a filter's impulse response and say one looks better than another. It is the way you asked it to be by changing the frequency and phase response. When combined with the speaker the impulse response will not look the same.

mptest-png.1048552


mptestgraph-png.1048553
 

Newman

Major Contributor
Joined
Jan 6, 2017
Messages
3,449
Likes
4,210
I reckon Dirac let the top octave roll off because it has a limit on safe or sane levels of gain to apply.
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
Here are some filters to try and some graphs to show the results with test convolutions and different windowing and smoothing. There are both 44.1 and 48K filters in the zip, there is a flat correction to add your own PEQ curve on and two target curves C2 and Cvar2.

The 5cycle FDW is close to the window size used in the processing and shows that the acoustic interference problems have not bee corrected but left alone. The phase is very close to minimum above 20Hz.

With Var smoothing, 1/48 or psychoacoustic smoothing things look much smoother and hit the target pretty well.

Above 10K there is rolloff in the flat file due to the gain being maxed out, but normally this is being dropped with a PEQ curve anyway. You could apply some post EQ there is you want with no real risk as the rest of the EQ is mainly reductive.

Will you like it? I have no idea :)
 

Attachments

  • pppsychsmoothtarget.png
    pppsychsmoothtarget.png
    30.8 KB · Views: 80
  • ppLprocVarsmooth.png
    ppLprocVarsmooth.png
    85.6 KB · Views: 68
  • ppLprocnowindow.png
    ppLprocnowindow.png
    99 KB · Views: 88
  • ppLproc5cycle.png
    ppLproc5cycle.png
    43 KB · Views: 96
  • pp-Toe-Flat-EP2.zip
    1.6 MB · Views: 55
OP
P

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,216
Likes
1,356
Location
Budapest
Thank you @fluid for creating these for me, I really appreciate them!
I have listened to all of them; for some reason they sound extremely bright, I would even say harsh
It is strange since on your graphs one could not tell that

If I load your last IR (highlighted) and the one that I am currently using I think it explains:
1652027087923.png


Mine is more 'conservative' above 400Hz
I really love playing around with these things in REW and RePhase - I have to say I really like the IR that I am currently using (attached)
Thank you to both of you @fluid and @ernestcarl for your kind guidance and advice

I have also decided to give Dirac another chance by applying some high shelf and low shelf preEQ upstream in the audio chain (so it will 'see' a somewhat corrected response: 'only' 20dB variance in the SPL instead of 40dB...)
I will report my findings here in a few days' time
 

Attachments

  • Current IR.zip
    127.8 KB · Views: 57
OP
P

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,216
Likes
1,356
Location
Budapest
I have listened to all of them; for some reason they sound extremely bright, I would even say harsh
Disregard this comment please, I just noticed that there were some additional filters turned on higher in the signal chain :facepalm::facepalm::facepalm:

Actually the filters sound fine! I will spend some time tomorrow listening to them
Thank you again!!
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
Actually the filters sound fine! I will spend some time tomorrow listening to them
Thank you again!!
You're welcome :) The filter without a target should sound very bright and is designed as a base to put your own PEQ tilt with. The others should give you an idea of some slightly different tilts.

My own preference now is to use the flat filter with a series of shelving filters in PEQ that I adjusted by ear to set my preferred balance. I would recommend trying this. I found the band around 150Hz and 1750Hz to be quite critical and very slight changes there gave a welcome change to the sound. The overall balance from bass to treble is important, whether it is boosted bass or dropped highs, if the slope is the same it sounds very much the same.

There is still an amount of alchemy going on with any in room measurements and EQ process to account for directivity and room differences so one size isn't going to fit all.
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
@fluid

I was wondering if it can be useful to apply a very steep low pass filter were one to use an enormous boost in the very HF?

1652493574192.png


Maybe (I'm just guessing) as to not excite potential "breakup" modes or long decay higher up. Hmmmn...
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
Usually it is a good idea to include some sort of upper HF rolloff to avoid creating a rectangular window between the filter and nyquist cutoff if there is boost. Running at a higher sampling rate would make more sense if you want to try equalization so high in frequency.
 
OP
P

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,216
Likes
1,356
Location
Budapest
Just an update on this topic: I have played around a lot with REW and RePhase and got really nice results at the end of the day! (correcting magnitude response and correcting phase response based on excess phase, etc.)
But then I realized that convolution introduces so much delay that video and audio get out of sync...:facepalm:
It is a non-issue when watching movies with Jriver but when I watch YouTube or anything else outside Jriver through its WDM driver the problem arises. Actually this is mentioned very clearly in Jriver's Wiki page but I still wanted to give it a try....won't work
Apparently I need to stick to solutions that do not involve convolution
I will post my results in due course
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Just an update on this topic: I have played around a lot with REW and RePhase and got really nice results at the end of the day! (correcting magnitude response and correcting phase response based on excess phase, etc.)
But then I realized that convolution introduces so much delay that video and audio get out of sync...:facepalm:
It is a non-issue when watching movies with Jriver but when I watch YouTube or anything else outside Jriver through its WDM driver the problem arises. Actually this is mentioned very clearly in Jriver's Wiki page but I still wanted to give it a try....won't work
Apparently I need to stick to solutions that do not involve convolution
I will post my results in due course

Did you try adjusting the buffer?

Audio options:
1660057504619.png



What's the time delay according to rePhase?

1660057387025.png



Cause even in my own various setups with 60ms FIR delay, lip sync is not an issue anymore with streaming video services e.g. Netflix, Disney+ and Youtube.

1660056927497.png



DSP Audio path:

1660056983574.png

*It takes a bit of time for the highlighted number to stabilize. And if stuttering occurs during live playback, a simple reset or clicking "stop" fixes it.

There really is no perceptible lip-sync issue if that number is kept below 200 ms with WDM/"Live" external audio streams -- at least with my cheapo 8ch DAC (still can't believe I only spent CAD $99 for it many years ago).


This can also be easily confirmed via some online AV sync tests:

1660057551078.png



Anyway, I think the key is to split/allocate your filters efficiently by fully utilizing JRiver's PEQ section rather than dumping everything into an FIR filter.
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
Remember it is the impulse peak position in the FIR filter that determines it’s latency (besides buffers) a very long FIR filter can have very low latency, but what you can’t have is a filter that manipulates phase at low frequencies and low latency.
 
Top Bottom