Hello, I write from Italy, I apologize for my English.
OCA i have done your guide for inverted convolution with great results.
I have now downloaded the latest version of rew 5.20.14 and redid the procedure with the updates mentioned in your new pdf.
In the inversion procedure now it is necessary to set the target on AUTO and not on 3.6 dB. The problem is that L CORRECTED is not aligned to the target but about 3 db above it! What am I doing wrong?
My correction is 55hz-1900hz.
Whereās R CORRECTED showing up? As long as they are at the same level thereās no problem. If thereās a difference, you can adjust SPL offset values of 1/A-MP according to the target level shown in the notes section of 1/A operations. Please note that the convolution impulse files generated do not have volume information in them.
Yes, no problem at all. REW uses a 3dB offset during some trace aritmethic operations to avoid division by 0 errros during fast fourier transformations. Testing the filters by ear is usually the easiest way to move forward.
Just to note that I unchecked Apply IR Window before Export as Mark Reed did and my bass was saved and itās brilliant. Iām not so sure Iāve done the best job but improved by at least 10%, probably more!
So, based on this info from Miska himself, if we plan on using this convolution filter in HQplayerā¦.should we not be exporting the convolution file in 192k sample rate if for example, we are streaming Qobuz or similar?
Thanks
Edit**** actually, I think the toggle in the convolution Settings that says āExpand HFā is there for this purpose.
I have not used HQPlayer myself but people here have successfully implemented these filters with it. You will need to export separate mono convolution files for the left and the right speaker at 48kHz.
First, thank you so much for spending the time and energy to post these videos and documentation.
I followed your directions exactly, but had a couple of issues along the way, mostly that my Overlay/Impulse screen doesnāt look anything like yours - i originally tried taking measurements using REW on MacOS, but switched over to Windows (using ASIO drivers @ 48K) to see if that would change anything and it did not. I do not see an acoustic Ref Peak (and my waveform looks much different), no matter how many times i try to take measurements. Iām using a UMIK-1 mic to take measurements, and iāve tried using both the output of my laptop, and an external USB audio interface with no difference. i skipped the Impulse part of your instructions and moved on to the rest and created Corrected curves etc, but iām not sure if that makes the rest of the workflow wonky or not. Once i loaded the convolution files into Roon i found that the system sounded a bit dull and lifeless compared to the non-convoluted output.
Not seeing acoustic reference chirp is normal. It only means your system doesnāt have any crosstalk between channnels which is in fact how a clean system should be. However, when I look at your measurements, I see over 8 ms system delays which corresponds to 2.5 meters. Thatās absurd, something is set up incorrectly in the measurement chain. There may be a bunch of reasons, some DACs take time to activate after the first signal arrives. I suggest you download the latest early access version of REW from the link below:
Use Java EXCL drivers to eliminate potential ASIO buffer size related problems, make sure USB Audio and Umik is configured correctly in Windows Sound settings (48kHz sampling rate, all enhancements disabled, exclusive control to other applications, etc.) and in REW tick āPlay dither before sweepā under options in the measurement window.
I donāt see any reason why this shouldnāt work. Once you have a descent impulse response for the left and right speaker, share them with me and I will prepare your Roon correction filters.
It would be interesting for me to see how Kanta 2s measure in other rooms.
Btw, I think the drop in frequency response after 10kHz in both of your speakers is possibly because your speakers are standing too low. Kanta 2s tweeter vertical dispersion is very narrow and they need to be at or above ear level. I had to place 10cm high blocks under the spikes to achieve this as my sofa was also a bit lower than standard. For similar reasons, they also like to be toed-in all the way towards the LP.
thanks for the response - i just updated my REW to 5.20.14, and re-did my measurements using Java EXCL. made sure things were configured correctly.
i actually do have my Kantaās raised a bit - i am using IsoAcoustics Gaia II speaker isolators below the speakers, which raise them a bit. they are right about my ear level when i sit on my couch.
Your Windows PC is connected to a DAC with USB and the DAC is connected to the Krell amp, rite? I donāt know what can be causing this delay in that chain but it should be an easy fix given its absurdity.
If the variations of the settings below didnāt work:
It could be about the input and/or filter settings in your DAC.
Itās a bit too late here in Germany but I will check tomorrow. Feel free to send measurements if you manage to get it right.
i plugged my laptop directly into the DAC via USB and it dropped my system delay measurements to ~ 1ms so it seems to be much better. (i was previously using analog outs of laptop, and a separate DAC, which seemed to be the cause of that delay).
attaching updated measurements here - there are 2 per channel:
Direct Roon convolution to this file and your system should sound noticably better. It might require 5-6dB headroom for clipping:
I am not very happy with these filters though because thereāre still fundamental problems with your measurements. Right speaker is always delaying more than the left. L and L2 have reversed polarities which shows an inconsistency between measurements. I had to pick the best ones out of each measurement because vector averaging simply nulls them. Is there like a speaker distance setting in the Dac or the amp? Or are all speaker cables correctly plugged in.The setup still needs investigation IMO.
Note: Try taking 1M length measurements, 4M might be too long for a USB mic. This is not the cause of the problem but always take 0-24000hz measurements, these were 20-24000.
thanks so much for this - the system def sounds better with this convolution file.
noted re: 0-24000hz - when i switched back to REW on my mac that must have gotten reset.
all interconnects and speakers cables are plugged in properly, in phase. and there are no distance settings on DAC or amp - amp and preamp are fully analog.
iām wondering if the measurement anomalies are related to the actual room layout - my LP is on a couch directly in front of the wall opposite the speakers which does create a bit of a bass trap. pics attached.
No, the different delays between channels are independent of the speaker or mic location. Whatever it is, itās in the hardware. Your room setup seems fine in general except for the open space to the rear of the left speaker which is causing less bass from that one but it can be fixed with DRC.
And I have checked all your measurement settings, nothing seems to be wrong at that front either!
When you can, try taking repeated (like 3 or 5 each) 256k long measurements for each speaker, your version of REW can do that automatically in one go.
For reference, this is my measurement info for the left speaker for a 1M length:
Holy moly those are big speakers in a small room ! (Not a criticism)
Sometimes when Iām listening to my speakers (in my small room) and trying to optimize my listening position, I think ā uh this is now ānear fieldā listening ā¦ and then ā¦ uh this is now āearspeakersā why donāt I just put on headphones