Part 2 - 2.1 Stereo Calibration - what am I missing?

[ddude003]

Keith_W, it is OK to agree to disagree... In the last few years traps and absorbers have added features like strike plates, diffusers and tuned membrane so as to not absorb mid and high frequencies...

My bad AOC vs OCA... I am dyslexic after dark... BTY AOC is a sociopathic socialist grifter bartender posing as a politician...

 

[Dave the Rave]

I did some more reading. Archimago's essay was quite revealing. I can see why its hard to correct HF unless we get excellent anechoic measurements of the speakers - something I am too novice to try. Hence his explanation to focus more below the Schroeder frequency range looks appealing.
Below are my tests lately.

1. Avoided any digital corrections above 500h. Just played around with speaker placement again and their mid/high knobs to achieve a flat response. Managed to have decent responses as per pic 1.

2. Since my speakers are 3-way, I wanted to re-evaluate them without the sub. After switching off the sub, the speakers LF measurements showed a flat response but astonishingly provided more coherent listening with the Mids and Highs. Realised I was stubbornly chasing for the Harman target curve cranking up the sub unnecessarily over the years.
However, I did find the speakers could not deliver the consistent LF energy for bass-heavy songs.
So I switched the sub back on and played with all sorts of settings ensuring I get minimal lift from the flat response the speakers provided. I did struggle with the mid-bass being affected when I tried getting better sub contribution.
I achieved the best listening experience when I toned down Bass in my amp (better result for the mid-bass), set my sub as sealed (less peaks and dips), sub at 180 phase and XO where speakers LF response suddenly dropped. See Pic 2 comparing previous and new L+R response below 500h.

3. Finally, I was wondering if I can improve on that using DSP. So I tried GSonic Convolution DSP. Tried both the full range and LF correction filters. From a measurement perspective, they looks good. Pic 3 shows the Left channel changes where essentially the LF is boosted and other ranges corrected marginally addressing the 500h dip with some smoothing above 500h (I used the minimal +3db boost).

Upon listening, the volume went down significantly as expected. The bass correction filters (below100h) was more pleasing. However, over time, I did find it too much hence I went back to setting my system without Convolution.
No disrespect to the advice given here in terms of Convolution success. But maybe since I re-discovered the default nice sound of the full range speakers, I just don't want to ruin it too much (or maybe I haven't mastered the right way to measure and create convolution filters)

Next step is obviously room treatment which is on my list in the longer term.

 

[user44455555]

However, over time, I did find it too much hence I went back to setting my system without Convolution.
No disrespect to the advice given here in terms of Convolution success. But maybe since I re-discovered the default nice sound of the full range speakers, I just don't want to ruin it too much (or maybe I haven't mastered the right way to measure and create convolution filters)

Next step is obviously room treatment which is on my list in the longer term.

the most problem is that your room is not symetrical. you get longer reverb from left side. you can try left and right channel db average so both channels get same EQ. then no phase problems between left and right can happen. use phase linear EQ in bass reduce transients. I think db average sound best

you have a door on left side. maybe you can open the door a little or more and measure and hear again. this work as a huge basstrap. You have maybe luck if it not disturb neighbours. also if you have a wardrobe on left side of the room you can try open it a little and measure and hear again. if it is large and have more depth it have much influnence on sound. also try the generator in REW to play sine tones. then press the note step button and hear the low frequencies. and hear music later. you can do with door open or not and compare too. most important is that it sound good for your ears.if you want not correct so much you can try out difftrent smoothings. try 1/3 then all eq settings are less sharp. I use 1/12 this is less sharp in bass as the Var you use.

I think in rew it is also usefull when can set a % value how much the EQ is used. I use mconvolution in DAW and i do not correct 100%. around 80% i correct. what measure and what hear is diffrent. if you see dips in your FR, play sinus with REW generator best with both speakers and hear how much it differ. play 1 note step above the dip frequency and 1 note step below dip and look how much you need boost so have the opinion it sound same loud. in bass range ears are not so much level sensitive as in other range.

 

[ddude003]

An interesting article... http://archimago.blogspot.com/2026/06/sound-absorption-ikea-mittzon-floor.html

 

[Dave the Rave]

Per this thread, before calling it a day, I decided to give Convolution a last red hot go [(thanks to @ddude003, @Keith_W and few others ). I realised i also needed a proper DSP processor hence invested quite some time in getting Camilladsp working in my chain (64bit processing at very low latency). Essentially, I now send 32bit PCM in Camilladsp and I get it rendered back to 24bit before feeding the separate player/dac.
I then tried RePhase (with the some tutorials' help) . The great thing about RePhase is that it allows correcting both the Frequency Response and Phase generating one single Convolution file. I applied those in my system and boy I am glad I did spend time learning.

I can finally say that the newfound Convolution did NOT take away transients or dynamics out of the music which I always had a beef with. Actually, it sounded more alive than normal PEQ. Music has more bite, hidden instruments reveal themselves, some songs feel more like live music, top frequencies way cleaner e.g snare drum reverb can be heard more distinctly and cymbal shimmer is more resolving. Must be the combined effect of CamillaDSP + Rephase correction + high bit-depth processing + Oppo 103D as the player (apparently has good power supply).

Below is my summary process:
[REW]
- Adjust LP measurements throwing away any messed up IR
- Apply necessary corrections in analogue domain - amp balance/bass and speaker mid/high to minimise digital corrections
- Adjust sub phase, level and crossover to get smoother transition from sub to main (they always play together)
- Take 3 position measurements then RMS Average to derive the PEQ corrections (kept minimal)
- Apply IR Windows - Tukey 0.25, Left 125, Ref Time unchanged, Right 500, 15 cycles
- Generate Minimal Phase
- Copy Excess Phase and export measurements with No Smoothing

[RePhase]
- Load Excess Phase measurements
- Apply main's crossovers. Yamaha NS-1000x are 500 hz and 6000hz both set at LR 12 Oct.
- Apply Phase EQ with all Q at 1.00 and below 90 deg (low frequencies with few degrees adjustments)
- Export filter with 65k tap , Extensive optimization to -100db

[REW/Rephase]
- Import RePhase filter in REW.
- Calculate result: Excess Phase x Filter
- Check if phase is moving closer to 0 deg
- If not, adjust Phase EQ in Rephase and check REW result iteratively
- Did not chase detailed flattening, correct wiggles or address major cliffs
- Below are the corrected Excess Phase I ended up with (the Left Phase EQ also shown as an example).


Since this is my first attempt at using RePhase, I wanted experienced opinions on my approach. Is there anything else I should do to further improve the Phase or the approach - i read conflicting opinions on the both fonts with some users not using the Excess Phase measurement or aiming a near flat (0 deg) Phase graph within surgical EQ.

Thanks

 

[Keith_W]

I would caution against correcting excess phase for high frequencies using measurements taken from the listening position, particularly if you have applied a 15 cycle FDW / no smoothing. That's a very high resolution correction. If you want to apply high-res EP corrections, you need an anechoic or quasi-anechoic measurement of the speaker. If you attempt to correct high freq EP from the listening position, you end up correcting reflections. Put that into your speaker, and there will be left-right phase anomalies. That may be a bit difficult to detect if you don't know what to listen for. Basically the phantom centre image might drift as it hits different frequency bands, so-called "phantom centre image shift".

Basically: for high frequencies, if you have a good quality anechoic / quasi-anechoic measurement of your speaker, you can go at it all you like (within reason). If you don't, then gentle correction is the name of the game - otherwise you end up correcting measurement artefacts and that's not good.

However, what you described IS the correct strategy for low frequency correction.

 

[ddude003]

It looks like you have learned a lot and come a long way in your journey to audio nirvana... You might consider using another target curve other than flat... Maybe this might help... http://archimago.blogspot.com/2026/07/room-target-curves-then-now-and-options.html...

You might also feed your REW room measurements into this newer FIR creation tool and see (hear) how that compares... https://github.com/VilhoValittu/DecayCore/releases/tag/v1.3.7

 

[Dave the Rave]

Thanks for the feedback. I will get a proper measurement for HF - i have a boom stand so will get the measurement done very close to each speaker at tweeter level. Will get 3 measurements for L and R. Is that OK for a quasi-anechoic measurement?
I tried DecayCore and it is indeed quite good - it does not seem to impact the sound signature/tone staying closest to the non-dsp version. It see it leaves the FR curvature untouched except applying broad correction even at full range. I will get another measurement done later. Should I get it done on the quasi-anechoic measurement and apply phase correction full range or measure LP only (3 times and rms average) or measure 3 positions and rms average them? I understand Vector averaging sometimes may not give proper phase info due to mostly due to minotr mathematical differences when changing position.

 

[Keith_W]

The measurement mic needs to be at least 2 baffle widths away from the tweeter to capture the baffle step. Most people place the mic at 1m. If you have an exceptionally wide baffle, the mic may need to be placed further away. The baffle step is a volume loss of 6dB that occurs over four octaves, its centre frequency can be calculated with f = 115824/W where W is the width in millimetres.

 

[Dave the Rave]

I did some new measurements - one from LP and another 1m from tweeter (hopefully it is quasi-anechoic as impulse was flat except one spike). I actually applied all analog level changes - balance, dialing speaker mid/high knobs and sub xo settings to get the responses I pasted below. Sound without any DSP is already good. As I had an added improvement with Rephase previously, my intent was to load the Excess Phase into Rephase - correct up to 200h using LP measurement then correct the rest using quasi-anechoic measurement.
However, check the phase responses below - showing from the listening position and 1m from speakers' tweeters
I don't understand why the Right channel's Phase response is as such (despite being readable from listening position). I tried importing in RePhase but hard to understand what to correct.

On the other hand, I tried DecayCore with multiple settings. I found the sound best in that order 1. Minimal 2. Asymetric 3. Mixed 4. Linear. Also tried with FIR (amplitude) corrections and without. The latter (Minimal Phase without FIR correction) gives the best rendition - more noticeable transcient and removal of subtle boominess in the bass region for some songs. However, I want to compare DecayCore filters with RePhase ones.

Would really appreciate if anyone could chime in to explain the 1m Right channel strange Phase response and also best settings for DecayCore.
Thanks.

 

[Keith_W]

Please upload the MDAT.

 

[Dave the Rave]

Thanks Keith. After some reading, hope I found the issue - shifted the Estimated IR Delays (results shown below). Please do review the measurements and will consider any general guidance on correction extent. Mdat attached.

 

[Keith_W]

This is the 1m measurement of your left speaker. There is a very loud reflection (arrowed) at 1.1ms / about 0.4m path-length difference. This is not a quasi-anechoic measurement, and I would not attempt to correct any high frequencies using this measurement.

Especially since:

... the huge forest of peaks in the MLP measurement (green) tells me that it's a pretty small and reflective room that you are measuring in.

You can work out where those peaks in the ETC are coming from. In my Google Drive, there is a presentation I gave recently for our local audio club: MAC 1. Go to Slide 50 onwards for a discussion on the ETC. From slide 68 onwards, I show how to predict the position of the reflection peak using a tape measure and some simple maths.

What is impressive is that you have managed to push the noise floor to -140dB to the signal through vector averaging. Well done!

If you want to correct your high frequencies with rePhase, you will need to take your speakers outside and elevate them. When you take your measurement, examine the impulse response and ETC and look for reflections. If you see any, the measurement is rejected and you will have to rethink your setup.

On the other hand, your MLP measurement is suitable for room correction as it is. I would probably stop correcting at about 150-200Hz.

 

[Dave the Rave]

@Keith_W That's an impressive preso - thanks for that. Read till slide 60 and I can clearly see the spikes/reflections issues. It confirms why i was not getting great DSP outcome when correcting for higher ranges. My Fs seems to be around 160h (room is open space and untreated - not ideal but trying to get the most out of it). I don't think I will venture correcting above 200-300h in RePhase. DecayCore also gives noticeable changes when correcting below 200h so will stick with that. Lessons learnt so far - Use speaker direct correction for mid to high frequencies (i.e correcting from measured graphs do not yield the best result for me and understand the importance of anechoic measurement here). Use DSP for LF corrections - will keep trying different options here.

 

[ddude003]

Plus 1 for looking at the diagnostics from DecayCore my fren... Nice bit of work @Keith_W[/USER ... Watching to see wher you will goto next [USER=63649]@Dave the Rave ...

 

[Dave the Rave]

I made some subtle changes in my system again. I always get surprised at improvements even when I thought this is the best my system has sounded so far - hence it's a never-ending tinkering love story.

1. First off, as i played more with Phase correction, I could see the limits.
Full range correction sound great but can be overly squeaky clean for some songs (concurs with @Keith_W comments). Low frequency correction is excellent but could not address the deep bass valleys in my situation.
So again moved the speakers a lot (as feasible). I don't have the ability to apply variable XO phase/ speaker delays in my amp - my Minidsp SHD xo/delays/dsp could easily be a solution, but I am not a fan of digital preamp hence it stays in the closet.
Result of bringing the sub closer to the LP in pic 1 below showing the net improvement in the bass area.... really happy with no DSP applied yet!!

2. Addressed the lowering of mid to high FR issue (pic 2).
Compared changing those in the analog domain (mid/high of speaker knobs) against digital PEQ in CamillaDSP.
Surprisingly digital PEQ sounds better!!
@ddude003 was right pushing me towards CamillaDSP which does not cause transcient/dynamics loss perception I was biased previously.
I also tried PEQ filters via ffmpeg and RePhase (Phase + PEQ convolution). Found I get the best result when setting IIR/Biquad PEQ directly in CamillaDSP before any Convolution DSP. So massive tick to CamillaDSP capabilities.
Note, I only applied cuts with no boosts and wide 0.7 -1 Qs for both left and right speakers.

3. Tried Rephase and DecayCore (pic 3) for Phase correction (pic 3).
In Rephase, got best result without Phase EQ and using a single linearizing at 500h. I also tried bringing the phase curve closer to 0 deg with different setting for crossover/box. Surprising again less Phase correction (i.e less slant towards 0deg) sounded better - so better resulting graphs does not equate to better sound!!!
Then tried DecayCore without FIR Correction (as I already bake the PEQ IIR in CamillaDSP before Convolution). Found the Minimal phase mode better than Mixed mode (Linear/Asymmetric are not too far)
Finally compared Rephase vs DecayCore based corrections. I think i like the DecayCore Minimal correction better but will listen more.

Hopefully this helps others in search of improvements and reducing bias towards DSP.

 

[ddude003]

I made some subtle changes in my system again. I always get surprised at improvements even when I thought this is the best my system has sounded so far - hence it's a never-ending tinkering love story.

1. First off, as i played more with Phase correction, I could see the limits.
Full range correction sound great but can be overly squeaky clean for some songs (concurs with @Keith_W comments). Low frequency correction is excellent but could not address the deep bass valleys in my situation.
So again moved the speakers a lot (as feasible). I don't have the ability to apply variable XO phase/ speaker delays in my amp - my Minidsp SHD xo/delays/dsp could easily be a solution, but I am not a fan of digital preamp hence it stays in the closet.
Result of bringing the sub closer to the LP in pic 1 below showing the net improvement in the bass area.... really happy with no DSP applied yet!!

2. Addressed the lowering of mid to high FR issue (pic 2).
Compared changing those in the analog domain (mid/high of speaker knobs) against digital PEQ in CamillaDSP.
Surprisingly digital PEQ sounds better!!
@ddude003 was right pushing me towards CamillaDSP which does not cause transcient/dynamics loss perception I was biased previously.
I also tried PEQ filters via ffmpeg and RePhase (Phase + PEQ convolution). Found I get the best result when setting IIR/Biquad PEQ directly in CamillaDSP before any Convolution DSP. So massive tick to CamillaDSP capabilities.
Note, I only applied cuts with no boosts and wide 0.7 -1 Qs for both left and right speakers.

3. Tried Rephase and DecayCore (pic 3) for Phase correction (pic 3).
In Rephase, got best result without Phase EQ and using a single linearizing at 500h. I also tried bringing the phase curve closer to 0 deg with different setting for crossover/box. Surprising again less Phase correction (i.e less slant towards 0deg) sounded better - so better resulting graphs does not equate to better sound!!!
Then tried DecayCore without FIR Correction (as I already bake the PEQ IIR in CamillaDSP before Convolution). Found the Minimal phase mode better than Mixed mode (Linear/Asymmetric are not too far)
Finally compared Rephase vs DecayCore based corrections. I think i like the DecayCore Minimal correction better but will listen more.

Hopefully this helps others in search of improvements and reducing bias towards DSP.


View attachment 97560

View attachment 97561

View attachment 97563

Some nice work there @Dave the Rave ... You are showing us that these different approaches to Room Correction Systems and Tools can get you about as close as you can get to Audio Nirvana and with the least expensive Room Correction things you can get from open solutions... Bravo my fren...