Optimizing Acourate Room+Speaker+Sub Correction

[ecwl]

First, I really have to thank @The Computer Audiophile for inspiring this project.

My subs are set to crossover to compensate my main bookshelf speakers (which I don’t use crossovers for transparency reasons). I have been using parametric EQ to address bass issues from 20-100Hz for years (and lately 20-300Hz). But in the midst of the pandemic, and because of our great leader’s wonderful experience with Audiolense’s convolution filters that @mitchco created for him, I decided to get Acourate and try to optimize my setup further. And I have to admit I learnt a few things. Obviously, YMMV. And you’ll probably disagree with me on a lot of this.

Before, I start, the main reason why I use Chord DACs and Chord M-Scaler is because I find the long tap length filters from Chord to really create realistic transients. I have found many people who drop by my local dealer can’t hear the differences, e.g. hand clapping, drum strikes, guitar plucks, cymbals, triangles. In fact, when I first got my Chord DAVE and compared it to Mojo, I can hear many improvements but just not the transients. The other thing I found Chord DAVE to shine at is soundstage depth if it’s in the recording, whereas I’ve heard a lot of DACs create and homogenized artificial soundstage width while shortening soundstage depth even if it’s in the recording.

Now despite most people feeling that parametric EQs are transparent, I have to disagree slightly as I often do find that there is always a very, very subtle loss in transparency and soundstage depth. However, because my bass issue is sufficiently significant (as with most real room systems), there is no question that parametric EQ of the bass peaks improves the sound dramatically compared to the minuscule loss in soundstage depth.

But for Acourate, it was quite easy to decide the target frequency response to mimick what I previously used for parametric EQ and the natural roll-off of the high frequency in my room (which happens to co-incident with EBU 3276. The real challenge I found was choosing the appropriate Phase Correction and Pre-ringing Correction. Since I sometimes do turn off my subwoofers at night and just listen to the bookshelves on their own, I also created separate filters for them. And the bottom line in order to understand how the filters affect the sound, I ended up creating more than 30 filters before reaching what to me, was optimal.

My major experience/findings were:

1. Convolution filters (at least via Roon) are even “less transparent” than parametric EQ. Not in the sense that I really noticed any loss of transparency to the music but I do notice a much more significant loss in soundstage depth. However, the phase correction when done optimally (or close to optimal) offers so many other benefits that once again this mild loss of soundstage depth to me was acceptable. That said, I suspect the soundstage depth change may or may not be audible in your system. 

2. While it is theoretically possible to use a long excess phase Frequency Dependent Window to alter the step response to mimick the the ideal, it is fairly easy to go overboard and introduce excessive ringing or group delays above 100Hz. Given my room, the speaker design, and ultimately, I already use “suboptimal” integration of subwoofer to speakers, there is only so much phase correction I can do before unpleasant artifacts get introduced to the system

3. So within the limits not introducing group delay or ringing artifacts, there is still lots of options to set for the length/width of low-frequency vs high frequency excess phase FDW. And since the group delay artifacts from excessive correction are often generated in my system between 100-1000Hz, I essentially have a choice to set the width of low-frequency and high frequency excess phase FDW the same, or I can set the low-frequency higher or the high-frequency higher before group delay artifacts come in. And what I found was that while setting low-frequency excess phase FDW higher does create more realistic and wholesome drum sounds and resonances, it is important to set high-frequency excess phase FDW higher in general because for me, that determines the accuracy of the transients which are highly audible in my system. Basically, if the high-frequency excess phase FDW is set too low for me, the whole system loses PRaT. Moreover, even though graphically I can correct pre-ringing in the step response by engaging in more pre-ringing correction, I also found that the correction it self also worsens the transient accuracy as I can hear them so engaging pre-ringing correction also always seems to lose PRaT for me. For reference, I would cross check by plugging my headphones directly into Chord DAVE to ensure the transient attacks are not an artifact of phase error.

4. Because the system is asymmetrically setup, I can usually do more correction on the right side than the left. But I find that sometimes with excessively long/wide excessive phase FDW correction on the right, I can actually hear that the same transients on the left don’t sound the same as the ones no the right. In fact, my favorite test track for this is Unsquare Dance from Further Time Out. So I find that I’m always writing down the Interaural Coherence Coefficient to see what the maximal value I can get between the two channels and then while I can push the right channel correction more, I don’t let the IACC drift off by more than 0.1%-0.2% optimal.

5. So in the end, in Acourate, the way I found the optimal settings for my system was to set Pre-ringing correction to 0/0, start with excess phase FDW at 1/3 on each channel and just slowly increase it by 0.1, e.g. 1.1/3.1, 1.2/3.2, etc. Until I get Group Delay >100Hz or pre-ringing. And then let’s say the optimal I can get on that channel is 1.4/3.4 for low vs high-frequency excess phase FDW, I would then try increasing the low-frequency setting by .1 to see if it’ll introduce Group delay >100Hz or pre-ringing. And then I would do the same for the other channel while looking at the IACC to see what is optimal and how far I can push the other channel. At least to my ears, this seems to be the optimal balance between trying to obtaining optimal transient attacks (which is why I got my Chord DACs in the first place) while getting as much low-frequency correction so that I can get improved bass resonance sound, e.g. drums that I hear in Copland’s Fanfare for the Common Man.

I know this approach seems a bit different from what’s been written here on forums. I definitely re-read over and over again what @mitchco wrote here and in his outstanding book and of course most of what I’m doing is recommended by him. It was just that the end optimization for me was to preserve the transient attack while trying to improve the phase response of the system. It’s been a long month of trial and error but I’m very happy with the sound I got. It definitely pushed my system beyond what I was able to achieve with parametric EQ alone. However, my experience also makes me wonder if it is truly possible to automate this type of process as I have previously found that Dirac has been a hit or miss in various systems at my dealer’s, almost always resulting in a loss of transient attack, and I wonder if it’s because the phase correction was excessive at times and not at other times. I know Uli the designer at Acourate has mentioned that his experience with phase correction is that it is very system dependent so it is likely that my method might not even work for somebody else’s system. However, I thought i would share my experience.

[ecwl]

1 hour ago, fds said:

Indeed, a great post. Motivated by the great articles by @mitchco and his book I got a Win PC to use Audiolense. It is certainly great but not as automatic as I had thought before starting to use it. Still it works wonders in the region below 400-500Hz in my room to the extend that some tracks - previously basically unlistenable - are quite fine now without excessive room treatments. However, so far it is a bit of give and take since some magic in the mids/heights gets lost when I use it. Also my subjective best filter generated so far creates some preringing in my case that I do not like to see. Thus, I am probably still very much at the beginning of the learning curve. I expect that the findings described by @ecwl will be very helpful for me ... 

Yes. So I find setting the phase correction to be a big challenge because if you look at Acourate, the recommended first trial for excessive phase FDW is 1.5/3 by Uli the designer/programmer. Whereas the default for Acourate is 3/2 and the manual even says you can go up to 5/3. @mitchco was mostly using 6/6 and @Archimago was pushing to at least 4/4 or 5/5 in his system and they all have pre-ringing correction engaged. But perhaps their systems and rooms are better behaved although you can see in one of @Archimago screen caps that there is a Group delay >100Hz at least. Initially, I did notice that a higher low-frequency phase correction is more audible initially because of the improved bass performance but at the expense of transient attacks/accuracy. In fact, I actually created two final filters, one has the low and high frequency phase correction the same, like @mitchco and @Archimago and I even engaged pre-ringing correction for that filter (it ended up being I think 2.7/2.7 on the left and 4.0/4.0 on the right and the right has pre-ringing correction and the left doesn’t because any higher setting just induces >100Hz Group Delay). My other filter ends up being 1.6/3.4 and 2.4/4.4 with no pre-ringing correction which has slightly less satisfying bass phase coherence but in my opinion much better transient response accuracy. At the end of the day, it just shows the fundamental deficits of my speakers and room acoustics. I think that’s why @mitchco and others recommends going from passive XO to active XO but I can’t see myself doing that any time soon. Also, I can’t afford 3 DAVEs and 3-M-scalers and a complicated computer audio setup to feed all 6 drivers.

Another thing I can’t figure out from reading the manual is to see the IACC in Audiolense. I think once @mitchco has his new edition of his book out, I’d buy it to read more about the newer softwares such as Audiolense and Dirac to see his opinion on them.

[ecwl]

41 minutes ago, The Computer Audiophile said:

Getting the absolute best out of my system and these apps required me to engage @mitchco at https://accuratesound.ca/. He’s the professional who eats, sleeps, and breathes this stuff. I attempted to read everything but I just couldn’t convince myself that wouldn’t cause more harm than good. 

While I do think @mitchco is the master at this, I think my experience has made me question: Is there an optimal convolution filter (assuming you've picked your favorite frequency response, e.g. EBU3276)? Mathematically and graphically, there may be an "optimal" convolution filter that generates the best looking step response possible for specific speakers and room. However, I'm wondering for those generating convolution filters at home, whether some like me would prioritize high-frequency phase response for better transient attacks while others would prioritize a more even low- vs high-frequency phase response correction. 

Unfortunately because of the pandemic, I have not invited my friends over to listen to the various filters I generated. I would be curious whether they would prefer the pure parametric EQ correction, my favorite convolution filter or the more even handed convolution filter I created. To my ears, each filter seems to involve a set of trade-offs amongst soundstage depth/transparency vs bass phase response accuracy (bass coherence/drum sounds) vs transient attack accuracy. But all of them, when set correctly are a significant improvement over uncorrected, uneven bass response.

[ecwl]

So @mitchco really is a godsend and a master in convolution filter generation.

After another 1.5 weeks of playing around with even many more filters and spreadsheets, I realized the filter experiments I had at the start of this post were truly suboptimal. I had learnt a few more things:

1) As @mitchco said a longer excess phase correction of at least 4 cycles in the low frequency makes the bass sound dramatically better and should really be the minimum setting

2) I was wrong in assuming that only increasing high frequency excess phase windowing would improve the transient/step response. I know understand that the step response is really dependent on the system and the low and high frequencies. Hence, it is possible to increase the low frequency windowing and still improve the transient/step response. In the end, once I optimized the low frequency, I just kept experimenting with different excess phase FDW back and forth and go into the step response and measure what the actual value is and wrote them down on a spreadsheet, e.g. trying 6.2/2.7 vs 6.1/2.8 vs 5.7/3 etc. I ended up selecting the filter setting that has the highest vertical spike in the step response

3) The other way indeed to restore the transient response was to only do partial frequency correction and I tried that too, up to 800Hz. The results with long excess phase FDW in the low frequency only were also quite amazing.

4) Ironically, if I were to start with convolution filtering, and listened to the partial correction first, I strongly suspect that would have been my preference. But because I have experimented with so many filters, I am really appreciating the full range frequency and phase correction for the system because there is something special about the additional phase coherence/high frequency phase correction.

All of this pointed out to me how poor most of my personal experiences with reference systems are. As a result, I have trained my ears to perhaps listen for the less important things or even the wrong things and assume that those sonic characteristics represent optimal sound. But by being able to do significantly more frequency and phase corrections, I’m realizing more what true reference sound means and what characteristics to really listen for.

The entire experience also made me a little weary about upgrading my speakers. Partly, it’s a recognition that I am somewhat limited by my room acoustics and configuration. Partly, it’s the concern that the upgrade to an even higher end speaker system might not be as dramatic once a proper convolution filter is in place. But I’m sure whatever I decide, I am still getting this new level of sonic performance.

[ecwl]

31 minutes ago, jamesg11 said:

~Macro question re above measurement-hearing-room discussion:

what is some recommended essential reading for arriving at optimal speaker placement (ie 2 mains + 2 subs)?

Jim Smith’s Get Better Sound

Because you would then know the most important part is actually optimizing your listening position first, if possible. 

And then once you’re done following his recommendations, I would say use a microphone and measure carefully because I don’t have Jim Smith’s ears so I rely on REW and microphone to sort through the bass issues. 

[ecwl]

8 hours ago, jamesg11 said:

presumably mitchco’s convolution services would cover this

I was under the impression the service is just to let you setup your system the way you want and then the convolution filter will be created for you. But do let us know about the whole process.

[ecwl]

I can’t believe I’ve gone back to tweaking my convolution filter. Been doing this intermittently since October 2020 as I can see from my own post. I was originally going to play around with a few target responses but ended up still preferring EBU3276.

But to get an optimal frequency response curve, I have to embrace a -6.7dB drop in volume. So while playing around with the settings, I created a filter that only causes a -2.7dB drop in volume by using less aggressive correction (mostly in the midrange/treble which is just my luck). And I have to admit, i cannot hear the subtle changes due to reduced corrections in the midrange/treble frequencies but I can actually hear better soundstage depths and a more 3D volume for the sound in my system.

So to me, in an ideal world, it’ll be nice to have a perfect room.

Otherwise, once you decide you need to use convolution filters you’re basically trading off:

1) Active crossover with recording engineer DACs vs passive crossover with audiophile DACs

2) More bass phase correction vs better transient response (although the trade-off is not always absolute)

3) More aggressive frequency response correction (which would require a bigger drop in gain) vs better soundstage depth

At some level there is no free lunch. On the other hand, any properly constructed filter would sound better than an uncorrected room response. Maybe it just comes down to preferences up to a point.