GSonic Reference – Free Stereo Room Correction Tool (Measurement + FIR Export)

[OCA]

However Gsonic's algorithms to prevent unintended artifacts is so sophisticated

It depends on what you mean by sophisticated. What really matters is how much you enjoy what you’re hearing. This algorithm is more accurate than others, but the MIMO version is far more advanced in terms of sophistication.

 

[Plompudu]

That's not the spec for crossing over a sub with speakers. The +10 dB boost applies specifically to the mono LFE track to enhance rare low-frequency effects that are unique to movie content. Subs work at equal volume level with speakers crossed over with them for all other content.

You're absolutely right, my bad for this badly defined example!
I should have just dropped talking about Dolby or made clear what I meant by saying "+10dB similar to the Dolby Spec" instead of making it sound like it's part of the Dolby Spec.

If you are using something like that as your target curve, you cannot expect it to be distortion free at very high volumes.

Yes, but by limiting the boost below 30Hz to less then 6dB I'm able to play as loud as I ever want to hear personally; with the current implementation I'm forced to listen a bit lower then what I would personally like.
I'll try a custom Target that cuts the range that is imo too boosted next week, if I've got time.

You ignored my suggestion to try a less bass heavy curve.

My bad if it seemed like I'm ignoring your suggestions and therefore seemed disrespectful!
I've already tried a few Targets and I like the tonality in the one with the boost the most for my usual listening.

I also don't know what gain setting you are using with which sub and why. Regardless, the tool aims to optimize for correct sound production not for maximum headroom. One can get always the best headroom with no filters. And FWIW Dirac ART boost default is +18dB adjustable to +24dB (including satellite speakers) and with very little selectivity at what it boosts.

If the aim is "correct sound production not for maximum headroom", why are you limiting it to 9dB?
That seems contradictory / like a arbitrary value to me, due to this example:

For example at low level listening I'd be fine with a boost higher then 9dB, if it resulted in less audible Nulls / smoother response.

 

[OCA]

Yes, but by limiting the boost below 30Hz to less then 6dB I'm able to play as loud as I ever want to hear personally

Now, this sounds more like wrong roll off frequency detection to me. Can you share your measurement and filter files?

 

[OCA]

why are you limiting it to 9dB?

The only limiting factor was system volume drop, didn't want filters to dim the overall volume by more than 10dB. Otherwise, the way boost is applied in the algo, there is no limit that can cause distortion.

 

[Plompudu]

Now, this sounds more like wrong roll off frequency detection to me. Can you share your measurement and filter files?

The Roll off frequency detection is difficult in my setup, since I've got massive room gain that extends my response down to ~3Hz with my Dual SVS SB-2000

Edit: fixed link
https://limewire.com/d/QbKpN#5zT7UYk9z0

The Measurements are done with a 14Hz Q1.23 -8dB Low Shelf already applied via my MiniDSP to the Subs; I use a UMIK-I (711-88), so accuracy below 10Hz is questionable and should likely be assumed higher, since the cal is flatted out below 10Hz instead of rolling off:

The only limiting factor was system volume drop, didn't want filters to dim the overall volume by more than 10dB. Otherwise, the way boost is applied in the algo, there is no limit that can cause distortion.

That's a fair reasoning, but I don't like that you arbitrarily choose the value without us giving the option to change it.
In a system with a high enough Bit-Depth should the digital volume reduction be no problem in most cases regarding digital noisefloor.

 

[yelloguy]

I tried this yesterday. The first thing I noticed was that the volume was lowered. I listen using raspberry pi to Focusrite which then splits the stereo signal 4 ways to the mains, and to two subs. I measured and time-aligned the subs to each other and both subs to mains using REW. I then added a couple of PEQ filters to smooth out the bass bumps. Finally I added three filters - a bass and a treble shelf filter, and a mid PK. This gives me a car like adjustment for tonality. I have a headroom of -2dB in CamillaDSP. This is how I was listening BEFORE trying GSonic.

Once I loaded the GSonic filters while playing music, the sound went down maybe (guessing) 8dB or so. I did not have that much headroom in my software volume control. So I had to add a gain of 4dB in CamillaDSP. I'm not sure if that is kosher or if it is introducing any clipping. Mind you all of my filters are NOT present in this version.

I noticed the imaging was much more precise with GSonic! The sound was softer too. Part of it is because the B&W aluminum dome tweeters that I listen to have a higher treble so the full bandwidth correction was playing a part. But to my ears (apparently used to the infamous B&W brightness) this was too much correction. I went back to my PEQ version of the file and lowered the volume even lower than the GSonic version. Yet I found that I preferred my own version because it seemed a little more 'open.' There was more 'sizzle' in that version that GSonic removed. This GSonic run was with all default settings.

I did not have time to do freq sweeps post setup yesterday. I will find time to do that today. And I will try to do the correction only up to 300Hz (thanks Mr. Schroeder!) and maybe try a different curve (more bass).

Thanks for creating this amazing tool!

 

[Plompudu]

The Roll off frequency detection is difficult in my setup, since I've got massive room gain that extends my response down to ~3Hz with my Dual SVS SB-2000

One Idea I've got on how to solve this is adding a room gain compensation step that subtracts X dB/oct below the room gain frequency (calculated using longest dimension of the room) to get closer to the real response of the Speaker / Subwoofer:

Example just to quickly visualize my thought:

 

[OCA]

The Roll off frequency detection is difficult in my setup, since I've got massive room gain that extends my response down to ~3Hz with my Dual SVS SB-2000

Edit: fixed link
https://limewire.com/d/QbKpN#5zT7UYk9z0

The Measurements are done with a 14Hz Q1.23 -8dB Low Shelf already applied via my MiniDSP to the Subs; I use a UMIK-I (711-88), so accuracy below 10Hz is questionable and should likely be assumed higher, since the cal is flatted out below 10Hz instead of rolling off:
View attachment 524210
View attachment 524211

That's a fair reasoning, but I don't like that you arbitrarily choose the value without us giving the option to change it.
In a system with a high enough Bit-Depth should the digital volume reduction be no problem in most cases regarding digital noisefloor.

You're using a zero‑tilt target with a +10 dB bass shelf, which does not reflect natural in‑room response. The subwoofer is rated to roll off at 20 Hz, yet your trace shows flat extension down to 4 Hz with no boundary gain peaks. This suggests a hidden miniDSP with non‑standard MSO settings, operating well outside the design assumptions of this tool and what I would consider sound DSP practice.

Under normal conditions, the algorithm would not apply maximum boost near a driver’s roll‑off nor would it boost dips that should not be boosted but what you’re feeding it has likely compromised the natural phase coherence.

Given the above, I don't see any compelling reason to add more complexity to the tool to accommodate this non‑standard use case. I will not allocate any more of my time on this one.

 

[OCA]

I tried this yesterday. The first thing I noticed was that the volume was lowered. I listen using raspberry pi to Focusrite which then splits the stereo signal 4 ways to the mains, and to two subs. I measured and time-aligned the subs to each other and both subs to mains using REW. I then added a couple of PEQ filters to smooth out the bass bumps. Finally I added three filters - a bass and a treble shelf filter, and a mid PK. This gives me a car like adjustment for tonality. I have a headroom of -2dB in CamillaDSP. This is how I was listening BEFORE trying GSonic.

Once I loaded the GSonic filters while playing music, the sound went down maybe (guessing) 8dB or so. I did not have that much headroom in my software volume control. So I had to add a gain of 4dB in CamillaDSP. I'm not sure if that is kosher or if it is introducing any clipping. Mind you all of my filters are NOT present in this version.

I noticed the imaging was much more precise with GSonic! The sound was softer too. Part of it is because the B&W aluminum dome tweeters that I listen to have a higher treble so the full bandwidth correction was playing a part. But to my ears (apparently used to the infamous B&W brightness) this was too much correction. I went back to my PEQ version of the file and lowered the volume even lower than the GSonic version. Yet I found that I preferred my own version because it seemed a little more 'open.' There was more 'sizzle' in that version that GSonic removed. This GSonic run was with all default settings.

I did not have time to do freq sweeps post setup yesterday. I will find time to do that today. And I will try to do the correction only up to 300Hz (thanks Mr. Schroeder!) and maybe try a different curve (more bass).

Thanks for creating this amazing tool!

There's a link to speaker/sub alignment tutorial earlier in this thread. Make sure you time align and corss over your sub(s) correctly first. ANd then do not apply any other eq or filters on your own. Let the tool handle the rest.

The filters are supposed to dim the overall volume by up to 10dB but they will never clip. For comparison, there are bypass filters which only apply the volume offset with zero filters.

 

[kharan]

You're using a zero‑tilt target with a +10 dB bass shelf, which does not reflect natural in‑room response. The subwoofer is rated to roll off at 20 Hz, yet your trace shows flat extension down to 4 Hz with no boundary gain peaks. This suggests a hidden miniDSP with non‑standard MSO settings, operating well outside the design assumptions of this tool and what I would consider sound DSP practice.

Under normal conditions, the algorithm would not apply maximum boost near a driver’s roll‑off nor would it boost dips that should not be boosted but what you’re feeding it has likely compromised the natural phase coherence.

Given the above, I don't see any compelling reason to add more complexity to the tool to accommodate this non‑standard use case. I will not allocate any more of my time on this one.

It's a bit hard to determine what a "natural roll-off" entails. Plenty of active speakers have some bass boost in the filtering.

Especially modern ones like Neumann or Genelecs, though both also come with their own Room Correction solution. It wouldn't surprise me if Dynaudio, Focal and other more HiFi oriented brand did the same with their actives.

 

[OCA]

It's a bit hard to determine what a "natural roll-off" entails. Plenty of active speakers have some bass boost in the filtering.

Especially modern ones like Neumann or Genelecs, though both also come with their own Room Correction solution. It wouldn't surprise me if Dynaudio, Focal and other more HiFi oriented brand did the same with their actives.

That's eq inside the speaker, not massively delayed boosted reflection wave chopped off with filters to a certain flat shape.

 

[kharan]

That's eq inside the speaker, not massively delayed boosted reflection wave chopped off with filters to a certain flat shape.

I think the consequence is the same, you don't necessarily want the room correction software to add boosts where there is no or very little headroom.

In my case, the 50 Hz boost is borderline okay because I don't have boosts lower in the frequency range. But despite being technically full-range speakers, my speakers are only capable of doing so at a limited SPL, they are big bookshelves but still bookshelves, with a limited internal volume.

And Gsonic detects a speaker roll-off at 20 Hz, meaning, I assume, that it would apply boosts down to that frequency if necessary. I personally don't think it's very safe.

 

[Plompudu]

You're using a zero‑tilt target with a +10 dB bass shelf, which does not reflect natural in‑room response.

I disagree, unless you only assume far field systems.

I maybe should have mentioned that I'm using my Speakers at just ~2m distance in a small well sealed room; that's one of the reasons why I've selected a Target with just a 0.45dB/oct tilt (200Hz - 20kHz) - a stronger tilt sounds unnaturally dark to me (especially due to the stronger drop above ~3kHz)

I think, that the selection of my Target seems fair, even when looking at the Estimated in room response:

The subwoofer is rated to roll off at 20 Hz, yet your trace shows flat extension down to 4 Hz with no boundary gain peaks. This suggests a hidden miniDSP with non‑standard MSO settings, operating well outside the design assumptions of this tool and what I would consider sound DSP practice.

Under normal conditions, the algorithm would not apply maximum boost near a driver’s roll‑off nor would it boost dips that should not be boosted but what you’re feeding it has likely compromised the natural phase coherence.

Given the above, I don't see any compelling reason to add more complexity to the tool to accommodate this non‑standard use case. I will not allocate any more of my time on this one.

The only DSP that should be applied in my measurements, is the build in EQ in the Subwoofer and the Low Shelf to reduce the Infrasonics due to having a roll off lower then the room gain frequency.

I can turn it off, but in that case I'd once again run into the problem I mentioned earlier:

I guess I'll wait till this bug is fixed and then retry it while checking if all EQ and FIR filters are turned off and I just used time alignment and crossovers.

 

[yelloguy]

There's a link to speaker/sub alignment tutorial earlier in this thread. Make sure you time align and corss over your sub(s) correctly first. ANd then do not apply any other eq or filters on your own. Let the tool handle the rest.

The filters are supposed to dim the overall volume by up to 10dB but they will never clip. For comparison, there are bypass filters which only apply the volume offset with zero filters.

Thanks for the tip about what the bypass filters are for. But ya the time align and cross over is carried over from my previous set up and there are no other filters stacked on top of GSonic.

However, the bypass would not allow me to compare my previous setup. For that, I will have to switch the yml (config) file in CamillaDSP. And that results in volume jump - unless, I add a negative gain in mine... hmm... I might do that! But first I need to build a new set of filters that go to 300Hz and use Not Dr Toole curve.

 

[john61ct]

A tunable dynamic LFE boost analogous to loudness control?

Would not need to be limited to mono input, applied across the stereo L/R signals, but only actually relevant to those output channels to trueSubs (by the user' bass management "prior to" Gsonic) whether stereo of mono

That's not the spec for crossing over a sub with speakers. The +10 dB boost applies specifically to the mono LFE track to enhance rare low-frequency effects that are unique to movie content. Subs work at equal volume level with speakers crossed over with them for all other content. If you are using something like that as your target curve, you cannot expect it to be distortion free at very high volumes. Such curves are only ever used for low volume listening.

 

[john61ct]

It depends on what you mean by sophisticated. What really matters is how much you enjoy what you’re hearing. This algorithm is more accurate than others, but the MIMO version is far more advanced in terms of sophistication.

I only used "sophisticated" as a (sincere) compliment to you. In my (limited) reading I get the impression "don't DRC above Schröder" is pretty standard advice wrt other tools.

From the fact that you only just added that as a user setting after being requested

I infer you were pretty confident that advice only apolies to Gsonic in certain limited scenarios.

 

[john61ct]

Make sure you time align and cross over your sub(s) correctly first. ANd then do not apply any other eq or filters on your own. Let the tool handle the rest.

Please allow me to dig a bit deeper, and ofc anyone could answer, I recognise all this is above my paygrade, maybe I'm totally off track?

I was thinking to test/attempt

A. "per-speaker as anechoic as possible" compensation EQ

with my mix of one-driver "active system" boxen vs passive multi-way speaker enclosures with internal crossovers

BEFORE then doing

B. MSO / REW + rePhase / miniDSP for crossovers & phase tuning / time alignment

Some say do not bother with step A at all, or maybe only do for the DIY MBM couplers not deep trueSubs, nor the main pair LS50s up top

What say you?

And what do you think about my setting independent bandpass boundaries per LF box, rather than true crossovers? Even if more difficult to blend, relying on REW iterations, I think would give more flexibility, especially for where overlaps might be useful?

 

[_giacomo_]

Yes, but only above room transient frequency...Below that it's "hyperbolic tangent" capped at 9dB with selective boost, correctly clip protected for any boost required. I don't intend to revise the very carefully calibrated algo unless you had an audibly bad experience with the results that requires revisiting. The +9dB cap of GSonic is not the same as boost limit of typical target matching least square methods.


No need to use any advanced settings (defaults to full range) unless you have electrostatic speakers.

Am I correct reading your sentence like this:
Not like least-squares methods:

Dirac-style / naive DRC -> can overboost like crazy
GSonic -> intentionally conservative

meaning GSonic won’t do silly, speaker-killing boosts chasing a flat graph.

If a dip is too deep, too narrow or clearly a cancellation, it likely won’t be boosted much (or at all) by GSonic.

I'll try ASAP. I can't allow a DRC filter to put extra strain on the (old and no more available as spare parts) Nautilus 801 woofers. That's my preoccupation.

Your work looks more than promising.

 

[Rja4000]

I released a new tool yesterday and wanted to share it here

Looks really interesting !
Thanks

Is there a way to limit corrected frequency range ?
It's very important to be able to set at least a high frequency limit, in my experience. Low limit may also be usefull.

341ms latency clearly exclude any studio use.
Do you have any plan to include a "low latency" option, using a combination of IIR and FIR ?

 

[Sir Sanders Zingmore]

thanks @OCA an amazing piece of work.

I have an observation

When I play the “test sound” it seems really quiet. The screen of my RME ADI-2 DAC shows something that doesn’t look at all like white noise.