[gammi]

This is something that i think you'll find interesting.

I literally have that video open and playing on my PC right now as I opened this thread

 

[jlaako]

The HQPlayer manual contains quite a bit on which filter & modulator *could* work well in which scenario. I have the impression though that the developers have put the "genre" recommendations based on speaker-based listening, which are not phase linear reproduction systems anyway (except maybe on full-range speakers? I digress). So on headphones I consistently look for linear phase filters that I think sound good, regardless of the manual recommendations.

It is not specific to whether you listen with loudspeakers or headphones. But more on signal properties of the source content, and how it was created. Minimum-phase filters are more natural, since they don't have pre-ringing / "pre-echo". But this is emphasized only on content with regular transients with high frequency content. Like rock for example.

 

[jlaako]

i got a feeling that pop/rock/electronic is generally recommended on filters that are technical worse, since these genres tend to mask errors somewhat with clipping, loadness, distortion etc
tho... i listen nearly everything BESIDE classic, just some very occaisonal film score and i still like technical correct filters... it makes a difference with the "right" rock/pop songs

No I recommend filters that are technically more correct for content that contains frequent transients. Classical music doesn't really contain such, so there's no need for filters to focus as much on time domain performance, than on rock music where the transient performance is much more important, since you have for example frequent cymbal drumming. And such recordings are usually made is acoustically dry multi-track studio environment where the only "space" is some synthetic reverb. While if you listen to organ music recorded in church acoustics everything in terms of the source content properties is different (no transients, but a lot of acoustic space)!

 

[jlaako]

I want to use the EQ profiles that I have for my different headphones that I already have on Roon and take them from there and use them on HQP directly but the UI is to byzantine for me as a neophyte to even understand what I'm seeing.

You can also just download parametric EQ .txt correction files from AutoEq page, and load that on pipelines 1 and 2 on matrix, enable the matrix and you can be happy.

If you want to compare, you can create matrix profiles which you can switch on the fly. For example empty profile for comparing against non-corrected case (you can also volume compensate this as necessary using the pipeline gain).

 

[jlaako]

maybe the display, i cant tell for sure.. but this "pattern" looks right for FIR, im just not sure about the amount...

Linear phase FIR filter always has some delay, from the beginning of the pre-ringing to the central main lobe. IOW, it has delay half of the filter's length. Delay equals to certain phase at given frequency. The graph just spins around 360 degrees (flips over between -180 and +180 degrees). In linear phase case delay itself in time is same for all frequencies. Minimum phase has minimal delay for all frequencies. Plot could be done in a different way too to remove the spinning, but I wanted to keep same approach for both cases.

if you use a two way speaker the tweeter is most probably wired out of phase... in this setup i advice to make sure that the bass driver is the right polarity and the tweeter is indeed inversed .... inverse phase is more noticable on bass drivers, it can really mess up the bass where highs merely sound a bit muted or a tad less real in comparison

Analog filters are minimum phase, for example speaker cross-overs. Doing such with a long (steep) linear-phase FIR would sound really bad due to the pre-ringing/pre-echo in the middle of audio band. You would easily hear transients having distinct "halo" around them.

Reason for having tweeter inverted depends on the cross-over and physical measures of the drivers when they are mounted on the same baffle. This is because distance difference of the acoustic centers (voice coil distance from the baffle). You can get around this with a coaxial driver, but such has other issues, because of reflections of tweeter sound from the moving bass/mid cone...

Also for headphone EQ, you may prefer to keep minimum-phase correction for filters that have some steeper (high-Q) corrections. Linear-phase version of such may begin to sound irritating.

 

[Ghoostknight]

In order to run DSD512, you need to be using ASIO drivers for your DAC, just make sure you have them install.

He mentions Fedora in signature. Linux supports native DSD too.

yep Linux, actually i switched to Debian a few months ago.. with Liquorix kernel, i might need some digging, one other issue i have with this setup currently: i cant play native dsd files... ... upsampling to DSD256 works, but .dsf files in DSD256 dont

As the 1st thing I would check if you have the latest nVidia drivers installed. Every new HQPlayer version is based on latest nVidia drivers and may crash with older driver.
Use the latest Studio driver, unless you are also playing games on the same computer, in such a case install the latest Game Ready driver.

yep latest nvidia drives are installed, for linux i think there is only one driver (if we dont count legacy drivers)
i might need some testing with the default kernel

i kinda hoped someone use Linux and knows this issue

After the driver update just try to compare PCM to DSD512 with and without CUDA, starting with some light filter like poly-sinc-short-mp-2s. Your CPU is OK for running any modulator, but your GPU may be too slow for processing filters at DSD512, resulting in CPU waiting for GPU to finish its job. If you don't get satisfying results, GPU upgrade is the way.

yea good idea, just to see if DSD512 works

I was using Sinc-Mx and ASDM7ECv3 last night.

this is was i currently use... and yes, i also have to wait 10 seconds or so till playback starts... sometimes the transition between songs arent perfect either
tho i only have intel 12400 + 1070 GTX

This is something that i think you'll find interesting.

yes, thanks

 

[jlaako]

Thanks for your thoughts! To me the NOS sound is very natural though a bit less detailed.

It also makes all transient timing / levels, and high frequencies broken. Because it lacks any proper reconstruction...

OS enhances detail but can lead to things feeling "smoothed over", I like to draw an analogy to vision how our eyes are extremely sensitive to bad image upsampling and reject overly blurry upsampled images as "vaseline smeared", sometimes visually you'd prefer a rough coarse but sharp image to a smooth blurry one.
Sinc-L is indeed mathematically the furthest thing from NOS except for one sonic trait, the ringing artifacts provide some sort of edge-enhancement that make it sound almost short again, see this nice visual description from Wikipedia:

This is bad example, because OS removes all those jagged edges. If you choose a suitably short filter, you don't have such ringing effects either.

I'm quite sure all interpolating filters keep the original samples intact, both polynomial and closed-form.

Also all halfband filters leave original samples untouched.

But please keep eye on the "Apod" counter! If it increments to figures higher than 10 during a track, there are many error components in the source data, so the original samples are wrong to begin with. Apodizing filters can fix these to large extent (not always completely though).

Since the concept of bit-perfect is out the window when upsampling to DSD my explorations of the ultrashort end of the filter spectrum led me to try minringFIR as a filter for DSD. In the end I settled on poly-sinc-mp as it has similar transient quality but is apodising. I previous discarded it since most people say minimum phase is bad for headphones (which are linear phase devices) and I thought it was superseded by the more efficient Gauss filters, but it sounds really good right now.
Not sure yet what I prefer yet between PCM-polynomial-2 and DSD-poly-sinc-short-mp. Depends on headphones and amp and music

How about poly-sinc-gauss-short?

 

[jlaako]

how does syncM (or better syncMGA) fit into this from tour perspective? i liked it a lot more than syncL but i might retest, tho my memory is blurred because i compared syncMGA

sinc-M / sinc-Mx is apodizing, as is sinc-MGa. These are overall technically better (higher attenuation, etc) than sinc-L(s/m/l), which are non-apodizing by definition, certainly interesting unique category, but require relatively huge length for decent performance.

From practical point of view, sinc-M is sort of variant of poly-sinc-ext3 while sinc-MG(a) is sort of variant of poly-sinc-gauss-xl(a).

apodizing errors seem to be something that is worth getting rid of from what i heared, even if we are all used to it from normal PCM playback, there are only a few filters that are apodizing

There's quite a number of apodizing filters, I believe at least about half of the filters. Please see the table in manual (or Help-page if you are using Embedded).

 

[Ghoostknight]

Also for headphone EQ, you may prefer to keep minimum-phase correction for filters that have some steeper (high-Q) corrections. Linear-phase version of such may begin to sound irritating.

yep highpass, lowpass can be quiet obvious in terms of preringing, thats also my advice, to avoid these if you use FIR, i noticed this with a additional subsonic highpass at 10Hz, while i sometimes prefer less rumbling (and less distortion) this messes things up a bit beside cutting subsonics with FIR

high Q isnt that important i think... the gain you apply is actually usually more phaseshift bending then the Q factor (of course you can make it worse with high q but the gain determines largely how large the phaseshift is, and therefore how much FIR needs to correct

In theory you can set your processing to IIR and look for the phaseshift... if you stay under lets say 45-90 degree phaseshift overall you should be good to go for FIR with minimal preringing...

No I recommend filters that are technically more correct for content that contains frequent transients. Classical music doesn't really contain such, so there's no need for filters to focus as much on time domain performance, than on rock music where the transient performance is much more important, since you have for example frequent cymbal drumming. And such recordings are usually made is acoustically dry multi-track studio environment where the only "space" is some synthetic reverb. While if you listen to organ music recorded in church acoustics everything in terms of the source content properties is different (no transients, but a lot of acoustic space)!

thanks for clarification

i might have a question here... i noticed that nearly all Poly filters have somewhat a similar sound but are quiet different to FIR or sincM variants, i kinda dislike them, is there something to Poly filters that make them "worse" to others?

 

[plumpudding2]

Thanks for the reply! again I love that you're interacting with your users on multiple forums.

It also makes all transient timing / levels, and high frequencies broken. Because it lacks any proper reconstruction...

I know! I'm just very curious whether there's something that makes up for the poor transient timing and high frequencies that makes it appealing to me, or whether my ear just doesn't seem care much about these things. I would expect music reproduction that has broken transients and high frequencies to also sound wrong to the ears.

This is bad example, because OS removes all those jagged edges. If you choose a suitably short filter, you don't have such ringing effects either.

Sorry maybe I should have clarified that indeed the top picture represents a very long filter and the bottom picture represents NOS. You do agree though that the top picture is kind of a visual representation of what happens to your music transient with a very long filter?

Also all halfband filters leave original samples untouched.

Thanks for clarifying that! I think it will be very nice for PCM upsampling hi-res then, will give it a try.

How about poly-sinc-gauss-short?

I really love gauss-long and gauss-xla but for some reason gauss-short has never really clicked with me.. do you like it yourself or prefer poly-sinc-short-mp?

 

[roderickvd]

Since the concept of bit-perfect is out the window when upsampling to DSD my explorations of the ultrashort end of the filter spectrum led me to try minringFIR as a filter for DSD. In the end I settled on poly-sinc-mp as it has similar transient quality but is apodising. I previous discarded it since most people say minimum phase is bad for headphones (which are linear phase devices) and I thought it was superseded by the more efficient Gauss filters, but it sounds really good right now.

Following the journey from "true NOS" down through the rabbit hole, is it not interesting how it leads further to OS paths? I thought about recommending minringFIR or any of the short filters, but did not as you could say it's leaving the NOS territory which I thought you were after.

My journey did kind of go like that: going from pure ignorance ("A DAC filter? What is that?") to audiophile NOS insistence ("yay analog") to discovering it's actually the short filters I keep coming back to. Although since your post I've put closed-form on for a few days now. It does have a special thing going for it - quite warm and spacious, I think.

songs with large numbers of apodizing errors (i actually found some that reach 50k apodizing errors in a few minutes) sound less like what would you expect if someone says "this sounds analog etc"

Oh that's amazing. Could you share a few examples? Would be keen to audition them on apodizing vs. non-apodizing filters.

Also all halfband filters leave original samples untouched.

Hi @jlaako nice to see you here too! So to check my understanding, the polynomial "filters" also leave them untouched? I thought that fitting them onto splines would also show some error compared to the original samples. (Whether keeping those original samples so perfectly is so desirable is another matter...)

How about poly-sinc-gauss-short?

As well as poly-sinc-mqa/mp3 even if it's not MQA or MP3. I like it a lot on 16/44.1 FLAC too even when I know it rolls off early.

 

[plumpudding2]

Following the journey from "true NOS" down through the rabbit hole, is it not interesting how it leads further to OS paths? I thought about recommending minringFIR or any of the short filters, but did not as you could say it's leaving the NOS territory which I thought you were after.

Yes! I love how I keep trying different filters, trying to see if the special quality they appear to have makes them the ideal tradeoff. And yes you're right those are moving away from NOS but were to me more about trying to create something close to the NOS feel on the DSD side of things. (As I mentioned before intuitively I don't believe the ultrashort interpolation filters make a lot of sense for DSD).

My journey did kind of go like that: going from pure ignorance ("A DAC filter? What is that?") to audiophile NOS insistence ("yay analog") to discovering it's actually the short filters I keep coming back to. Although since your post I've put closed-form on for a few days now. It does have a special thing going for it - quite warm and spacious, I think.

It took me a long time to warm up to the short filters as it is easy to fall into the trap of "Long filters costs more cpu horsepower so they must be better right?". Also short filters make no sense from a classical sampling theory perspective. Most posts on the subject of filtering on ASR for example will tell you that any sort of impulse response ringing or other effects from filtering are inaudible and modern dac chip filters are as close to perfect as they need to be. I've been slowly piecing together why exactly that doesn't appear to be true in reality and dream of proving it mathematically one day

 

[Ghoostknight]

Oh that's amazing. Could you share a few examples? Would be keen to audition them on apodizing vs. non-apodizing filters.

Rolling Stones - Flip the Switch (remaster) - 9k apodizing errors
Dj Duro - Shizzle my dizzle (original mix) - 9k apodizing errors on half the song (somewhat a clean mix, not great for testing but "usable" for specific things imo)

more i cant find right now with large number of errors

i can somewhat hear the effect of the errors even with apodizing filters but the effect is largly reduced compared to non apodizing, im not sure if im right here tho or if other things like distortion blurr my perception... you mostly find large numbers of apodizing errors with distorting songs so its kinda hard to say

 

[bogi]

In the end I settled on poly-sinc-mp as it has similar transient quality but is apodising. I previous discarded it since most people say minimum phase is bad for headphones

poly-sinc-mp is high quality middle lenght apodizing minimal phase filter, which is very universally usable, perhaps except of classical and other acoustic music, where linear phase filters are more suitable. Sonically similar non apodizing middle length filter is for example poly-sinc-hb-m.

minring FIR is short rather lower quality filter but is very light on processing.

In one of my toddays posts I suggested some filters to try - from every category (long, middle length, short).

the polynomial "filters" also leave them untouched?

I mentioned it already today ... see filter table in HQPlayer PDF document, the last column. Starts on page 25. You can see polynomial filetrs are not apodizing so yes, they preserve original sample values. In the manual you find also filter description - they don't produce perceivable pre- or post-rignging, but their reconstruction quality is low and thus Miska marekd them in teh manual as not recommended. It is worth to open the manual, particulatly the tables of filters or modulators (or dither in PCM case).

 

[wagninger]

Hey folks,

I am currently stumbling over many online discussions about which hardware to get to run HQPlayer at pretty high settings optimally.

I am in a bit of an interesting situation right now in that I’m buying a “new” PC for the sole purpose of running HQplayer, Roon will run somewhere else.

The PC is from a friend of mine, so certain components are what they are, but I can choose the graphics card:

Ryzen 9 5900x (12 core, 3.7GHz)
32gb RAM
1TB m.2 Samsung 980 Pro
750W Corsair PSU

Now the question: for DSD 512 with the more demanding filters, should I go for an RTX 3080 10GB
or is an 8GB 3070 enough? Does anybody have experience with those cards for CUDA offloading?

Cheers!