1 hour ago, AccuMagi said:

With further digging into what Sean has done, it appears he applied the filters into the matrix rather than Convolution…, blah blah…, whatever that is! 

 

Appreciate the clarification of the differences. My zero knowledge of this subject matter told me that convolution needs sample rates to be in PCM. If that’s correct;

1. What happens to the DSD source/material? 

2.  Will it degrade the sq from multiple conversions from one format to another, then apply the filter? 
 

Cheers. 

 

I do have a license for an older version of HQP though. In HQP there are two places you can apply the convolution filters. I don't know the advantage of applying it in matrix rather than convolution, but there is no sonic difference either way. My understanding is that one allows you to easily switch filters, while the other doesn't. 

 

DSD material can remain DSD, or it can be converted to PCM. It depends on what settings were chosen. 

 

I believe there is no sonic penalty from converting from DSD to PCM for processing. The only penalty is using more CPU, and on my older version of HQP, and with 8 channels, that is a LOT OF CPU. I mean, enough CPU usage to bring my older i7-6700K to its knees and start thermal throttling and suffering from audio dropout. People who run HQP at its highest settings tend to use gamer type PC's with nVidia graphics cards to take advantage of CUDA offloading to process at the highest resolution possible. I have read descriptions of people using these monster watercooled PC's, placing them in other rooms to suppress the noise, and streaming the output to a HQP endpoint in the listening room. I haven't investigated doing that and I probably wouldn't consider doing it, because it to me the benefit of doing it are marginal compared to the gains you can get by refining the DSP. 

 

1 hour ago, frednork said:

Hi @AccuMagi,

this is a Hq player based question, Keith doesnt use it. Best to talk to Sean about how he has set up the filters but short answer (at least previously) was that DSD is converted to a psuedo pcm file to allow filtering and then converted back. As far as I understand it, it is not possible to convolve DSD directly but its not conversion to "normal" pcm either.

 

I thought that it was possible to convolve in DSD with HQP? Anyway, like you said, best to speak to a real HQP expert. My knowledge is out of date. 

32 minutes ago, Keith_W said:

I thought that it was possible to convolve in DSD with HQP? Anyway, like you said, best to speak to a real HQP expert. My knowledge is out of date. 

It has always been possible to do this with DSD.

 

People just heard the quote that "it is not possible", and so that will be repeated until the end of time.

The reason why it is/was "not possible" is because it was too computationally intensive for audio workstations of the 90/00s to do, practically almost at all, but definitely not in the real time (or close to) that audio workstation users needed.

 

In theory there could be a loss in quality... but in practise with high quality modulators and high sampling rates and precisions (hence the computation intensiveness) it is zero.    Fred is right that "internally" you could view what is happening as a sort of "pseudo pcm" .... in so far as the sample amplitudes are given finer (16bit, 24bit, 64bit, whatever) quantisation levels (so they can be adjusted), and then a "conversion to dsd" is run on the result.   But because the original sampling rate is preserved (or run even higher) any time domain issues in the conversion (which is where all the issues are) are avoided, and because it is all "offline" (ie. there is no real, imperfect, clock) then it's all just a maths, precision problem (which can be hand waved away by using oodles of precision).

On 7/8/2023 at 6:13 PM, Keith_W said:

I had a discussion with Uli about a new driver linearization technique that also flattens phase by convolving an All Pass filter into the correction. I believe that this method has not been described anywhere online.

Do you mean the method in Accourate ... or inverse all pass filter for correction of a speaker crossover in general?

 

If the later, then it's as old as the hills.    The phase response of a typical crossover is some variant of an all pass filter (depending on the XO shape) ... and so the phase linearisation is an inverse all pass filter.

There are talks of the target DSP curves to replicate specific speakers maker’s tonality. 

My love affair with the Marten Mingus Orchestra, Marten Coltrane Supreme 2, Kharma Exquisite Grand 3.0 and Tidal Audio La Assoluta come closer, do you think? 

20 minutes ago, AccuMagi said:

There are talks of the target DSP curves to replicate specific speakers maker’s tonality. 

My love affair with the Marten Mingus Orchestra, Marten Coltrane Supreme 2, Kharma Exquisite Grand 3.0 and Tidal Audio La Assoluta come closer, do you think? 

Yeah, that would be great. press the button and get any of those. You may be able to make them sound a bit more like them but there are things that cannot be improved by dsp which are baked in to the speaker design, such as  what the drivers are able to do, the design of the cabinet and crossovers. Can sure get the most out of your Magico's in your room, May still end up better in some/many aspects than those others mega speakers unfiltered though!

@AccuMagi I believe that you can exceed any of those speakers with an active DSP system. There are many advantages that simply can not be replicated in a passive analogue crossover, or if they can, can not be done to the same precision. You have to start off with a very good speaker in the first place, and I believe that of all high end speakers, Magico's are objectively the best designed. Like I mentioned in our chat, they own a Klippel scanner. I don't think anybody in Australia has a Klippel, and I would wager that it is very rare among boutique manufacturers. I am 100% sure that large companies like B&W, Focal, KEF, and definitely Harman own Klippels, but I have never liked the tuning of B&W's and Focals. 

 

Come visit to the SNA show and visit me in Melbourne. I can show you what can be done with a few pushes of a button. Tuning of other speakers can be replicated, if you know the target curve. As @frednork mentions, it won't be exact, but you can come fairly close. 

11 hours ago, AccuMagi said:

With further digging into what Sean has done, it appears he applied the filters into the matrix rather than Convolution…, blah blah…, whatever that is! 

 

Appreciate the clarification of the differences. My zero knowledge of this subject matter told me that convolution needs sample rates to be in PCM. If that’s correct;

1. What happens to the DSD source/material? 

2.  Will it degrade the sq from multiple conversions from one format to another, then apply the filter? 
 

Cheers. 

 

Hi @AccuMagi - it was fun listening to those special speakers. Very happy we got a good result this time!

 

Regarding, convolution vs pipeline matrix. The former is fixed and only works for basic room correction, with a fixed channel mapping.

 

Pipeline matrix allows up to 128 channels and all kinds of advanced routing options. And practically unlimited number of 'profiles' that you can switch 'on the fly' as music plays.

 

In short, nearly everyone will want to use the pipeline matrix.

 

When we eventually convert your speakers into the world's first DSP active Magicos  😀 (each driver with its own DAC channel and own amp channel and crossover inside HQPlayer), then the channel mapping options of Pipeline Matrix will be more important than now, with your systems 2ch config.

 

@davewantsmoore covered the convolution of DSD music files perfectly.

 

I would add - to do convolution of DSD tracks and output even higher DSD rate, you will actually need to get your nice NVidia 4070 GPU actually hooked up and working with HQPlayer. It will (I am guessing) be too much for your CPU alone. You can try that now and report back

 

For PCM content, as you know, no problems.

 

The GPU will also be essential for 8ch convolution when you have the world's first DSP active Magicos up and running 😄

 

 

2 minutes ago, rand129678 said:

When we eventually convert your speakers into the world's first DSP active Magicos  😀 (each driver with its own DAC channel and own amp channel and crossover inside HQPlayer),

 

So you will be removing the internal passive XO?

 

12 minutes ago, andyr said:

 

So you will be removing the internal passive XO?

 

 

I was partially joking. It would be cool for me to do but have not yet discussed it with the owner.

 

Something to discussed much later.

 

Edited August 11, 20232 yr by rand129678

Don't discuss it with him then Next time you are there, just rip out the passive XO! 

1 hour ago, rand129678 said:

active

At the risk of taking the half joke serious (but I think it has relevance to this thread) ....  Do you think you could improve the crossover(s) in the magico ?

Keeping in mind the the layout of the drivers on the cabinet, the directivity of the drivers, and the crossover filter shapes will have all been optimised as a group..... applying new filter shapes, even if we think they are "better" on some other metrics, will almost certainly upset this balance.

It may be that those things are not optimised well.  I have not seen individual driver responses from any magico speaker (aside from stereophile but this is not nearly enough data) ... but I would expect it is done quite well, given the price.

 

If not, then applying DSP to the "whole speaker" is the much wiser choice.

Don't get me wrong, I build exclusively active speakers and would not usually like to do it any other way.... but making a passive speaker into active, is of practically no benefit .... unless there are fundamental issues with the crossover components, driver/speaker impedances, or filter shapes.... and if it's the later, then this comes with a deep rabbit hole of pitfalls (driver layouts, directivity, and filters are all designed together)

I am listening again to what Sean did with the filter. The more I hear, the better my appreciation gets—more in-control sound stage and imaging. The sound image articulates exceptionally well. 

 

Accompany by excellent grape juice certainly helps!!🥳

@aris and @sir sanders zingmore came around last night and I learnt a thing or two about my ISO226 volume control. We had an interesting discussion. 

 

ISO226 is an international standard for equal loudness. Fletcher and Munson noted that as volumes get lower, levels of bass (especially) and treble needs to be boosted for the signal to be perceived to have the same tonality. It looks like this: 

 

 

I have two programs in the signal chain that can implement an ISO226 volume control via DSP - JRiver and Acourate Convolver. However, I set Acourate Convolver's volume to maximum (so the equal loudness curve is not applied), and use JRiver's volume control for convenience, because it can be controlled via a remote which is integrated in my playback app. 

 

When they came over and started playing their music, it sounded awful - much too bass heavy. I didn't notice because I was making tea and was not in the main listening position, and when I sat down in the other sofa I thought "hmm, there is wayyy too much bass, maybe the seat to seat variation in my bass is not as good as I thought". Then Aris complained there was too much bass and asked me for a filter which had less. I was surprised, because this was not a problem I had noticed before - but as soon as I sat in the main listening position, I realized he was right. 

 

After some agonizing as to whether I had gotten something wrong with my tuning (after all, the system measures as I had designed it - i.e. to follow the target curve which is supposed to sound balanced), I remembered that I was using the ISO226 volume control. As soon as we turned it off, everything sounded balanced. 

 

What is interesting is that when I am listening to my music (which is 100% classical), I strongly prefer the effect of the ISO226 volume control. I can turn the volume down to my usual low listening volumes and the system maintains its clarity and tonality. But with other types of music, it sounded awful. 

 

My theory is: the zero point of the ISO226 volume control determines when JRiver will start applying the curve. However, recordings are mastered at different volumes, especially pop music where (thanks to the loudness wars), the overall volume is much higher. Nearly all classical music is mastered at a much lower volume. This means, for a given volume setting, classical music will have much less ISO226 correction applied to it than pop music. If you look at the graph posted above, you can see that it applies a stupendous amount of bass boost if you turn the volume very low - this was causing pop music to sound incredibly bass heavy and thick. 

 

After they left, I made a more concerted effort to tune the zero point of my volume control and I still prefer it the way I had left it. However, it easily became too much if I started playing anything which was not classical. 

 

So, two lessons were learnt last night: 

 

1. If you use an ISO226 volume control, make sure you tune the zero point or it will make your system sound unbalanced, and 

2. GTG's are important because different types of music will make you realize deficiencies in your system that you were previously unaware of. I did not realize this with my earlier GTG this week (with @andyr and @Steff) because we were playing classical music during that GTG. 

Edited August 16, 20232 yr by Keith_W

I use loudness filters via CamillaDSP, configured to be active until a certain volume. Tuning what level is a matter of taste and depending on what levels you set the low and high gains to be. 
 

GTG and demonstrations are critical for system tuning! Different people, different music, and in my case different venues bring out the gremlins!

Edited August 16, 20232 yr by aris

I've been putting together a mostly non-classical music playlist of about 30-45 minutes duration for whenever we can arrange another get together.

Edited August 16, 20232 yr by Steff

@Keith_W make sure you have some Taylor Swift in the mix, you know you secretly want to, now you can claim that it's just for "educational purposes".

3 hours ago, Keith_W said:

@aris and @sir sanders zingmore came around last night and I learnt a thing or two about my ISO226 volume control. We had an interesting discussion. 

 

ISO226 is an international standard for equal loudness. Fletcher and Munson noted that as volumes get lower, levels of bass (especially) and treble needs to be boosted for the signal to be perceived to have the same tonality. It looks like this: 

 

 

I have two programs in the signal chain that can implement an ISO226 volume control via DSP - JRiver and Acourate Convolver. However, I set Acourate Convolver's volume to maximum (so the equal loudness curve is not applied), and use JRiver's volume control for convenience, because it can be controlled via a remote which is integrated in my playback app. 

 

When they came over and started playing their music, it sounded awful - much too bass heavy. I didn't notice because I was making tea and was not in the main listening position, and when I sat down in the other sofa I thought "hmm, there is wayyy too much bass, maybe the seat to seat variation in my bass is not as good as I thought". Then Aris complained there was too much bass and asked me for a filter which had less. I was surprised, because this was not a problem I had noticed before - but as soon as I sat in the main listening position, I realized he was right. 

 

After some agonizing as to whether I had gotten something wrong with my tuning (after all, the system measures as I had designed it - i.e. to follow the target curve which is supposed to sound balanced), I remembered that I was using the ISO226 volume control. As soon as we turned it off, everything sounded balanced. 

 

What is interesting is that when I am listening to my music (which is 100% classical), I strongly prefer the effect of the ISO226 volume control. I can turn the volume down to my usual low listening volumes and the system maintains its clarity and tonality. But with other types of music, it sounded awful. 

 

My theory is: the zero point of the ISO226 volume control determines when JRiver will start applying the curve. However, recordings are mastered at different volumes, especially pop music where (thanks to the loudness wars), the overall volume is much higher. Nearly all classical music is mastered at a much lower volume. This means, for a given volume setting, classical music will have much less ISO226 correction applied to it than pop music. If you look at the graph posted above, you can see that it applies a stupendous amount of bass boost if you turn the volume very low - this was causing pop music to sound incredibly bass heavy and thick. 

 

After they left, I made a more concerted effort to tune the zero point of my volume control and I still prefer it the way I had left it. However, it easily became too much if I started playing anything which was not classical. 

 

So, two lessons were learnt last night: 

 

1. If you use an ISO226 volume control, make sure you tune the zero point or it will make your system sound unbalanced, and 

2. GTG's are important because different types of music will make you realize deficiencies in your system that you were previously unaware of. I did not realize this with my earlier GTG this week (with @andyr and @Steff) because we were playing classical music during that GTG. 

 

A few things, if you are using JRivers Loudness feature the developers have stated on their forum it is not strictly ISO 266 but is based on it.

 

One of the most important documentations regarding JRiver and Volume.

https://wiki.jriver.com/index.php/Volume

 

Have you done the Reference Level Calibration as in the link above?

For Loudness to work appropriately it should be done although there is no harm done adjusting to taste by adjusting the Internal Volume Reference Level.

 

You should also use JRiver's Volume Leveling feature which will typically reduces the volume of loud modern pop music much more than classical and thus you will strike more of a balance between the two when using JRiver's Loudness feature. Because, it then won't overcompensate for loud recordings because you've turned them down so much more than typical quieter classical recordings. And the more you reduce the volume, the more JRiver's Loudness will be applied.

https://wiki.jriver.com/index.php/Volume_Leveling

 

I believe I have my system sorted to a point that all musical genres (at typically different loudness levels) work OK with JRiver's Loudness feature.

Edited August 16, 20232 yr by Satanica

Oh and I'm not sure if you know but you can choose to use Jriver's Loudness with Full Spectrum (default), Bass Only or Treble Only as of a couple of years ago and this can be set per zone.

 

NEW: Added the option to loudness to be full spectrum (default), bass only, or treble only.

 

https://yabb.jriver.com/interact/index.php/topic,129421.msg898268.html#msg898268

 

 

Edited August 16, 20232 yr by Satanica

3 hours ago, Keith_W said:

@aris and @sir sanders zingmore came around last night and I learnt a thing or two about my ISO226 volume control. We had an interesting discussion.

 

3 hours ago, Keith_W said:

2. GTG's are important because different types of music will make you realize deficiencies in your system that you were previously unaware of. I did not realize this with my earlier GTG this week (with @andyr and @Steff) because we were playing classical music during that GTG. 

 

1 hour ago, aris said:

Keith_W make sure you have some Taylor Swift in the mix, you know you secretly want to, now you can claim that it's just for "educational purposes".

 

1. Brilliant. Would have liked to be a fly on the wall for that meeting of the minds.

2. The benefits of GTGs!! ...... apart from the company, the food, and the beverages.

3. "You know you want to". Excellent.

Such a good thread. 

2 hours ago, aris said:

@Keith_W make sure you have some Taylor Swift in the mix, you know you secretly want to, now you can claim that it's just for "educational purposes".

 

I DON'T WANT TO!!!!!!! 

 

I felt dirty and violated after you played that song! I had to clean out my ears with some Bach afterwards!! 

 

1 hour ago, Satanica said:

A few things, if you are using JRivers Loudness feature the developers have stated on their forum it is not strictly ISO 266 but is based on it.

 

One of the most important documentations regarding JRiver and Volume.

https://wiki.jriver.com/index.php/Volume

 

Have you done the Reference Level Calibration as in the link above?

 

Thank you for your links. I was not aware they had updated their wiki on loudness. I have been using this feature since MC25 (or whatever it was when they debuted it) and did an initial calibration with an SPL meter and by ear. I have left it alone since. In fact, I have used it for so long that I usually forget that I have it turned on, it's only when I run measurements using JRiver as a convolver that I look at the curve and think "that's weird..." before remembering I have loudness on and turn it off. I think it's time to revisit that calibration. I'll do it later today. 

 

@Steff you are welcome any time, just send me an SMS when you would like to come. 

The other thing everybody agreed on was that this target curve sounded the best: 

 

 

I had a number of target curves loaded and we were able to switch between them with Acourate Convolver. It's a rather strange target curve with a slight 0.5dB bass boost between 50-80Hz, a pivot at 200Hz, -5dB at 10kHz, and -10dB at 21khz. I haven't seen this type of target curve before. It's not my target curve, it's Uli's. It sounded more full and had more body, but to me it lacks a bit of top end. Up till recently, my preferred target curve was a Harman-like curve, with a 3dB bass shelf to 100Hz, slowly tapering to 200Hz, then -4dB at 21khz. This is a neutral sounding curve which I like because it offers so much clarity, mostly because there is more treble energy. I am going to use this curve as a basis for further experimentation. 

1 hour ago, Keith_W said:

I DON'T WANT TO!!!!!!! 

 

I felt dirty and violated after you played that song! I had to clean out my ears with some Bach afterwards!!

I don't think I would have wanted to either.

 

Some Julia Lezhnova singing Handel or Bach (either JS or JC) would give your speakers a decent workout, and any of those would probably ease your troubled ears (from the earlier experience).   

Edited August 16, 20232 yr by parrasaw

1 hour ago, Keith_W said:

Thank you for your links. I was not aware they had updated their wiki on loudness. I have been using this feature since MC25 (or whatever it was when they debuted it) and did an initial calibration with an SPL meter and by ear. I have left it alone since. In fact, I have used it for so long that I usually forget that I have it turned on, it's only when I run measurements using JRiver as a convolver that I look at the curve and think "that's weird..." before remembering I have loudness on and turn it off. I think it's time to revisit that calibration. I'll do it later today. 

 

That's good to calibrate to -20dB relative to a full-scale.

But, as I wrote before you need Volume Leveling -23dB relative to full-scale plus 3dB (use EQ) to match above for Loudness to be functioning well consistently.

Edited August 16, 20232 yr by Satanica

9 hours ago, Satanica said:

A few things

I miss JRMC sometimes...

7 hours ago, Keith_W said:

I haven't seen this type of target curve before. 

Look pretty standard if you squint at it. 

 

If you "delete" the extra roll off above 10Khz, then it's the standard ....   ~ -4dB straight ish slope from 1khz to 20khz .... and then below 1khz there's a a boost of ~ a few dB.   In this case the LF boost begins up high (right up a 1khz)... but it's only ~1dB / octave.

 

7 hours ago, Keith_W said:

, it's Uli's. It sounded more full and had more body, but to me it lacks a bit of top end. Up till recently, my preferred target curve was a Harman-like curve, with a 3dB bass shelf to 100Hz, slowly tapering to 200Hz, then -4dB at 21khz. This is a neutral sounding curve which I like because it offers so much clarity, mostly because there is more treble energy. I am going to use this curve as a basis for further experimentation. 

👍