convolution room correction and room curve

linuxonly · Aug 22, 2024

moedra said:
How do your filters sound now?

First, louder. I gained 9 dB using that method, What a surprise.
Then, the stereo image is perfect. Bass response is more linear (sounds like e.g. Supertramp, School). Voice is clear, natural and well centered, Instruments sound like actually instruments and are easily spotted.
And the latency dropped so much that one can now watch videos using this filter.
From now on, I can't see how anything could be added to get it to sound better.

There is a psychoacoustic smoothing in REW. I didn't play with it yet.

Thanks a lot. That is amazing work, a piece of art!

moedra · Aug 23, 2024

JStewart said:
Thank you for this! I’ll give it a go first chance I get.

Questions...

I see your measurements start at 0Hz. Any concerns or advice on potential woofer damage? @linuxonly measurements start at 20Hz. Is this also acceptable?

Would you see any value to applying different smoothing to different parts of the frequency range. 1/3 octave smoothing above the “transition“ zone, for example.
Using your method and a bit of copy and paste to exported files, I think this could be accomplished.

No. That noise is not coming from my speakers. They drop out just below 40Hz. That sub response there is noise from the interface getting nabbed by the mic, if anything. The target is designed so that the filter gets rid of anything there anyway so it's alright.

Perhaps there is value in it. Personally I like the idea of Var smoothing, but the level of smoothing is too low in the highs and I don't like it. If I could have 1/12 in the lows and 1/6 in the highs I'd like it better. Acourate's psychoacoustic FDW of 15 cycles low and 15 cycles high is similar to REW's 1/12 smoothing.

Perhaps John could give us a smoothing blender for the Var type. Something that would allow us to choose a smoothing level for the two ends of the frequency spectrum and a way to control the frequencies at which the smoothing amounts start. For instance, when you look at Audiolense's Correction Procedure Designer, you are given a high and a low frequency and you can choose the FDW at each one of those frequencies. The area between is interpolated from one point to the other. I would love it if we could customize Var smoothing this way.

moedra · Aug 23, 2024

linuxonly said:
First, louder. I gained 9 dB using that method, What a surprise.
Then, the stereo image is perfect. Bass response is more linear (sounds like e.g. Supertramp, School). Voice is clear, natural and well centered, Instruments sound like actually instruments and are easily spotted.
And the latency dropped so much that one can now watch videos using this filter.
From now on, I can't see how anything could be added to get it to sound better.

There is a psychoacoustic smoothing in REW. I didn't play with it yet.

Thanks a lot. That is amazing work, a piece of art!

Thanks. I have been working on this subject for a couple of years. Personally I don't think psychoacoustic smoothing is detailed enough in the bass. Maybe that is a personal preference. I think 1/12 smoothing sounds great, but as I said in my response to JStewart I wish we could customize the Var smoothing option. That way we could dial in 1/12 low and psychoacoustic high, for example, if we wanted to.

moedra · Aug 24, 2024

Hey guys, I put the process into a tutorial. Take a look and share it if you know anyone who wants step by step instructions.

Digital Room Correction with REW | Meridian Soundworks

nverxion.wixsite.com

ddude003 · Aug 24, 2024

Nice work guys... And thanks for sharing your tutorial with everyone @moedra...

JStewart · Aug 25, 2024

moedra said:
Hey guys, I put the process into a tutorial. Take a look and share it if you know anyone who wants step by step instructions.

Digital Room Correction with REW | Meridian Soundworks

nverxion.wixsite.com

Very well done!

linuxonly · Aug 25, 2024

moedra said:
Hey guys, I put the process into a tutorial. Take a look and share it if you know anyone who wants step by step instructions.

Digital Room Correction with REW | Meridian Soundworks

nverxion.wixsite.com

Super!

bixite · Aug 29, 2024

moedra said:
Hey guys, I put the process into a tutorial. Take a look and share it if you know anyone who wants step by step instructions...

Hey there, I am doing FIR filter experiments for the last two months now and have tried out many different approaches - some found on the internet, some adapted, some own ideas. I am in the process of changing my preferred workflow basically once a week

... but it gets more and more elaborated.
Currently, I am mainly working with REW. I also tried out automatic filter creation with DRC (plugged into FusionDSP which is Volumio plugin).
Now, this thread and your work really inspired me! A huge Thank You for the fantastic approach and your documentation.
I found out the following:

Minimum Phase on the target does not work. When target and measurement are both minimum phase, the A/B arithmetic of REW leads to a result without an impulse response. This might be due to changes in the software (I don't know).
I found a way to build "mixed-smoothed" measurement instead of using the 1/12 smoothed data. Brand new, just today, it works, but I did have no time for extensive testing and listening. Basically, I am starting with 1/24 in the bass, moving to 1/12 in the mids up to 2kHz, switching to 1/6 from 2kHz up to 5kHz and using 1/3 smoothing for 5kHz and above.

I would be highly interested in some exchange.

Thanks again, kind regards and greetings from Hamburg
Bidjan

linuxonly · Aug 29, 2024

bixite said:
Minimum Phase on the target does not work. When target and measurement are both minimum phase, the A/B arithmetic of REW leads to a result without an impulse response. This might be due to changes in the software (I don't know).

Hi,

What do you mean by that? I tested this on 3 systems and here are my results. Did you make a MP version of the A/B result?

Capture d’écran du 2024-08-29 09-34-12.png

Capture d’écran du 2024-08-29 09-24-54.png

Capture d’écran du 2024-08-29 09-23-47.png

bixite · Aug 29, 2024

linuxonly said:
Hi,

What do you mean by that? I tested this on 3 systems and here are my results. Did you make a MP version of the A/B result?

View attachment 73286 View attachment 73287 View attachment 73288

Thanks for the quick reply and the right question. I missed to generate a minimum phase version of the A/B results. I just tried out and now I have an impulse response. I will try out and check the differences (I achieved an impulse response when using the non-MP-target/L-MP.

It works and sounds good. I will do some measurements and more listening. When I generate minimum phase on the A/B results, does it still make sense to export the minimum phase version of the IR? Isn't this applying minimum phase to an already minimum phase filter?

linuxonly · Aug 29, 2024

bixite said:
Thanks for the quick reply and the right question. I missed to generate a minimum phase version of the A/B results. I just tried out and now I have an impulse response. I will try out and check the differences (I achieved an impulse response when using the non-MP-target/L-MP.

You're welcome. Glad you did.

bixite said:
It works and sounds good. I will do some measurements and more listening. When I generate minimum phase on the A/B results, does it still make sense to export the minimum phase version of the IR? Isn't this applying minimum phase to an already minimum phase filter?

Maybe you got a point. It does no harm though. IMHO it's just the export wav settings are undocumented. So going for the MP version seems the proper thing to do.
You could try to reimport the IR wav files and multiply them with the reimported original response MP to see whether you get MP results, you should

Capture d’écran du 2024-08-29 10-27-57.png

Capture d’écran du 2024-08-29 10-27-28.png

bixite · Aug 29, 2024

linuxonly said:
You're welcome. Glad you did.

Maybe you got a point. It does no harm though. IMHO it's just the export wav settings are undocumented. So going for the MP version seems the proper thing to do.
You could try to reimport the IR wav files and multiply them with the reimported original response MP to see whether you get MP results, you should

View attachment 73289 View attachment 73290

I will take some time to compare different variants with some math and listening.

bixite · Aug 29, 2024

bixite said:
I will take some time to compare different variants with some math and listening.

I did some homework. I created 3 filter pairs. All used as input smoothed minimum phase frequency responses for L and R and a target curve with the same goal parameters in two versions (non MP and MP).

v1: Target-noMP/L-smoothed-MP (same for R), then Export as minimum phase IR (I did not generate MP-versions of the filters)
v2: Target-MP/L-smoothed-MP, then Export as minimum phase IR (I generated MP-versions of the filters since they do not have an IR after A/B)
v3: Target-MP/L-smoothed-MP, then Export as IR (I generated MP-versions of the filters since they do not have an IR after A/B)

Same for the right channel. Results so far:

Math: After Reimporting the filters and applying them to the frequency responses v1 provided smoother results than v2 and v2.

Measurements: After measuring v1, v2, v3 (L+R, L, R), v1 also looks better. The frequency response is closer to the target. The phase in the bass region looks better, the step response settles quicker. v2 and v3 are very similar to each other.

Listening: Just started, v1 is my current favourite, v2 and v3 are extremely similar

Any comments, hints very welcome!

P.S.: More testing required! My v1 does not smooth the way I wanted. I get 1/48 results where I expected 1/6 results. The MP-target seems to work differently in the trace arithmetic. I looked up the REW documentation:

"The result of arithmetic on measurements that have compatible impulse responses is smoothed using the measurement A smoothing, unsmoothed data is used during the calculations. Other measurements use whatever smoothing they already had applied during the calculations and the result is treated as unsmoothed (or 1/48 octave smoothed if data is 96 PPO)."

I need to do more testing here. I get different results with differently smoothed Targets (at least visually in REW).

JStewart · Aug 29, 2024

bixite said:
I found a way to build "mixed-smoothed" measurement instead of using the 1/12 smoothed data. Brand new, just today, it works, but I did have no time for extensive testing and listening. Basically, I am starting with 1/24 in the bass, moving to 1/12 in the mids up to 2kHz, switching to 1/6 from 2kHz up to 5kHz and using 1/3 smoothing for 5kHz and above.

Could you share your workflow to accomplish this?

moedra · Aug 29, 2024

bixite said:
Hey there, I am doing FIR filter experiments for the last two months now and have tried out many different approaches - some found on the internet, some adapted, some own ideas. I am in the process of changing my preferred workflow basically once a week ... but it gets more and more elaborated.
Currently, I am mainly working with REW. I also tried out automatic filter creation with DRC (plugged into FusionDSP which is Volumio plugin).
Now, this thread and your work really inspired me! A huge Thank You for the fantastic approach and your documentation.
I found out the following:

Minimum Phase on the target does not work. When target and measurement are both minimum phase, the A/B arithmetic of REW leads to a result without an impulse response. This might be due to changes in the software (I don't know).

I found a way to build "mixed-smoothed" measurement instead of using the 1/12 smoothed data. Brand new, just today, it works, but I did have no time for extensive testing and listening. Basically, I am starting with 1/24 in the bass, moving to 1/12 in the mids up to 2kHz, switching to 1/6 from 2kHz up to 5kHz and using 1/3 smoothing for 5kHz and above.

I would be highly interested in some exchange.

Thanks again, kind regards and greetings from Hamburg
Bidjan

Did you follow the tutorial step by step verbatim here? I typed that up as I created a filter and it all works great and sounds amazing. However, if I missed something or made a mistake, let me know and I'll fix it.

moedra · Aug 29, 2024

bixite said:
I did some homework. I created 3 filter pairs. All used as input smoothed minimum phase frequency responses for L and R and a target curve with the same goal parameters in two versions (non MP and MP).

v1: Target-noMP/L-smoothed-MP (same for R), then Export as minimum phase IR (I did not generate MP-versions of the filters)

v2: Target-MP/L-smoothed-MP, then Export as minimum phase IR (I generated MP-versions of the filters since they do not have an IR after A/B)

v3: Target-MP/L-smoothed-MP, then Export as IR (I generated MP-versions of the filters since they do not have an IR after A/B)

Same for the right channel. Results so far:

Math: After Reimporting the filters and applying them to the frequency responses v1 provided smoother results than v2 and v2.

Measurements: After measuring v1, v2, v3 (L+R, L, R), v1 also looks better. The frequency response is closer to the target. The phase in the bass region looks better, the step response settles quicker. v2 and v3 are very similar to each other.

Listening: Just started, v1 is my current favourite, v2 and v3 are extremely similar

Any comments, hints very welcome!

I remember during testing I had created filters with the raw no-phase target and it did not sound as good as the one made with a minimum phase target. Yet I am still interesting in hearing experiences from other people and perhaps improving the guide. I need to know specifics, though, so write down steps and changes you happen to make.

Having said that, however, I am comparing the filter made with my guide to a minimum phase filter created by Acourate. The two sound virtually indistinguishable from one another except for a miniscule (maybe 2%) difference in the target slope. It's hard to imagine beating Acourate.

bixite · Aug 29, 2024

JStewart said:
Could you share your workflow to accomplish this?

I tried out a lot of functionality within REW and found no way. But then came to an easy manual approach. I exported the measurement with different smoothings. From the exports (all .txt files) I took out the respective portions, combined them into a new .txt file and then imported this new file. Now, I need an expert who tells me which smoothing to use for which frequency span

moedra · Aug 29, 2024

linuxonly said:
You're welcome. Glad you did.

Maybe you got a point. It does no harm though. IMHO it's just the export wav settings are undocumented. So going for the MP version seems the proper thing to do.
You could try to reimport the IR wav files and multiply them with the reimported original response MP to see whether you get MP results, you should

View attachment 73289 View attachment 73290

In my tests, leaving the inverted responses as they were and exporting the wav with the min phase version radio button produced the best results. This is how I wrote it in the guide. Let me know if anything you find makes an improvement and I'll try it out myself. If it's just an extra step that doesn't make a difference, though, let's leave it out.

moedra · Aug 29, 2024

bixite said:
I tried out a lot of functionality within REW and found no way. But then came to an easy manual approach. I exported the measurement with different smoothings. From the exports (all .txt files) I took out the respective portions, combined them into a new .txt file and then imported this new file. Now, I need an expert who tells me which smoothing to use for which frequency span

I would use 1/12 low and 1/6 high. Put the switch somewhere around your room's transition zone.

I still think it would be great if John included a customizable Var smoothing model with interpolation.

bixite · Aug 29, 2024

moedra said:
Did you follow the tutorial step by step verbatim here? I typed that up as I created a filter and it all works great and sounds amazing. However, if I missed something or made a mistake, let me know and I'll fix it.

Thanks for your reply. I followed your procedure. I stumbled upon the MP version of the target. When doing A/B with Target-MP/Smoothed-L-MP my result has no IR and thus cannot be exported as IR. Did you generate MP-versions of the filters? If so, why do you export an MP-version of an MP-filter? Is this not applying MP twice?

My workaround was to use the target without minimum phase. Then A/B with Target/Smoothed-L-MP generates a result with an IR. I just found out that this is NOT working the way I want because it applies the smoothing of the target to the result (which was 1/48). Now I could smooth the target, e.g. 1/12 , and the result would be 1/12 smoothed. Interestingly, the result looks different when the Target is minimum phase and the smoothing of B seems to play a role now ...??? I have to understand REW's trace arithmetic better ...

bixite · Aug 29, 2024

moedra said:
In my tests, leaving the inverted responses as they were and exporting the wav with the min phase version radio button produced the best results. This is how I wrote it in the guide. Let me know if anything you find makes an improvement and I'll try it out myself. If it's just an extra step that doesn't make a difference, though, let's leave it out.

see my other reply - with A/B using Target-MP and Smoothed-L-MP the result has no IR (using REW 5.40 Beta 46)

bixite · Aug 29, 2024

moedra said:
I remember during testing I had created filters with the raw no-phase target and it did not sound as good as the one made with a minimum phase target. Yet I am still interesting in hearing experiences from other people and perhaps improving the guide. I need to know specifics, though, so write down steps and changes you happen to make.

Having said that, however, I am comparing the filter made with my guide to a minimum phase filter created by Acourate. The two sound virtually indistinguishable from one another except for a miniscule (maybe 2%) difference in the target slope. It's hard to imagine beating Acourate.

I will do more testing tomorrow and over the weekend and let you know. Afer some listening v1 is not really better but a little sharper in the highs and that is because my intended smoothing has not been applied. I will make more attempts with using the MP-Target but at the moment do not get an IR result with A/B ... maybe I am missing something.

linuxonly · Aug 29, 2024

moedra said:
In my tests, leaving the inverted responses as they were and exporting the wav with the min phase version radio button produced the best results. This is how I wrote it in the guide. Let me know if anything you find makes an improvement and I'll try it out myself. If it's just an extra step that doesn't make a difference, though, let's leave it out.

Happy to clarify.

I did no test anything outside your method. I followed your method Target-MP / L Smoothed-MP (A/B = L Inv). Exported as wav with the MP radio button selected.

Everything else I said is only theory to determine why the poster had different IR than myself.

moedra · Aug 29, 2024

bixite said:
Thanks for your reply. I followed your procedure. I stumbled upon the MP version of the target. When doing A/B with Target-MP/Smoothed-L-MP my result has no IR and thus cannot be exported as IR. Did you generate MP-versions of the filters? If so, why do you export an MP-version of an MP-filter? Is this not applying MP twice?

My workaround was to use the target without minimum phase. Then A/B with Target/Smoothed-L-MP generates a result with an IR. I just found out that this is NOT working the way I want because it applies the smoothing of the target to the result (which was 1/48). Now I could smooth the target, e.g. 1/12 , and the result would be 1/12 smoothed. Interestingly, the result looks different when the Target is minimum phase and the smoothing of B seems to play a role now ...??? I have to understand REW's trace arithmetic better ...

No, I didn't convert the inversions. I found that redundant as well. They are left as-is after the division, which is how I wrote the guide. The end result sounded better by doing it that way.

With regard to the target and smoothing, the responses are exported with 1/12 smoothing. When you bring them back in, even if they are divided by a 1/48 target, the data is still 1/12 because that is all the responses contain. I think you may need to update your version of REW. Use the newest beta and see if your experience is any different.

moedra · Aug 29, 2024

bixite said:
see my other reply - with A/B using Target-MP and Smoothed-L-MP the result has no IR (using REW 5.40 Beta 46)

Something seems wrong there. When I do it, everything works fine. If it didn't, I wouldn't have written the guide that way.

convolution room correction and room curve

Member

Member

Member

Member

Senior AV Addict

Senior AV Addict

Member

Member

Member

Member

Member

Member

Member

Senior AV Addict

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Popular tags