HQPTuner: an improved configuration interface for HQPlayer Embedded

Same track:


My fault for using a poor example the second time around.

Actually, seems to be maybe a mix? Some of that is definitely images and some of it is definitely cut off but not everything.

Also, the “brickwall” (a somewhat generous term for a small bump) can be seen in the average over time from the first image:

I’m not seeing images there… But to me that looks like quite common hires recording. Not knowing the content, there can be always digital synths recorded through their analog output, and in such case you have bandlimited output and commonly also some images from the not so great internal DAC of the instrument.

But you also get pretty steep roll-off with large diaphragm microphones. Many multitrack recordings are mixture of different types of microphones.

That SMPS tone tells it is not an upsample, but instead recorded from analog domain in 96k.

Very interesting.

The content is Tool’s “Schism”, by the way. No digital synths or anything that should have been band-limited. Another shot from the same band that is 100% obviously real from a different album:

I suppose that also means this one, that I’ve been calling fake, is also real? Amnesty (I) by Crystal Castles:

That is definitely a band-limited synth or similar, as you say.

Tool is familiar band, I have their Fear Inoculum album.

Real in a sense that it is not an upsample. It could be mix made on analog console, from 44.1k or 48k recording, recorded in 96k from the analog mixing console outputs. Not a true hires at least on that position, since there’s nothing to fill up the higher sampling rate. But not “fake” in the worst sense. It is a bit like someone recording a solo piano at 96k sampling rate, whereas the instrument produces frequencies only up to about 16 kHz… Not particularly useful…

2 Likes

Just to say first that I appreciate this discussion and your wisdom.

Let me make sure I have this right… because if I’m thinking about this correctly, Amnesty (I) is real but would still benefit from the 20k filter: Amnesty’s sampler ran through a brickwall reconstruction filter too, ringing captured by ADC. So… same damage, same fix?

The problem I’m running into is that this rule…

…doesn’t stand on its own. See Marked for Death’s spectrum above, totally black above 22.05k. It’s fake, but there are no images to be seen. (Or maybe they’re technically there but just so faint that they’re not visible?) Then look at Amnesty (I) and pretend there’s no spur.

To an algorithm, these are identical: brickwalled with effectively nothing up top. But one is fake and one is real. I’m not seeing how I can distinguish them.

That said, if the damage is the same and the 20k filter is beneficial either way, that issue dissolves.

It doesn’t look particularly steep, so no need there. But it doesn’t harm either, at least you get rid of the SMPS whine.

Yes, it has gone through upsampling filter that has enough attenuation.

We have also seen albums that have been upsampled without any filter, resulting in full level images.

And then gazillion that are somewhere in between, with various levels of images.

Damage is not necessarily the same. But you can run 20k filter systematically for content upsampled from 1x source. Then there are those 2x, 4x, 8x filters that address upsamples of corresponding sources (96k etc).

1 Like

I don’t find Schism from Qobuz, so I cannot quickly check it out.

But for example this is a real thing:

P.S. Better run these through apodizing filter!

It’s there, the album is Lateralus, track 5.

I know :sob: Rage Against the Machine’s self-titled album remastered at 24/192 is real and generally clean, except that there’s about 2,000 apod events per second. It drives the counter crazy like nothing else in my library.


The tiniest little bump around ~31k can be seen in the median… I personally wouldn’t engage the junk filter over that, but I’ve been wrong about this before :sweat_smile:

That is clean as-is. I wouldn’t do anything to that one.

I see Apod 10—is that the reason? Otherwise, it sounds like true HiRes to me, though with very high energy in the audible frequency range.

Yes, the material is such that it triggers the apod detector. poly-sinc-gauss-hires-lp used here takes a good care of it.

1 Like

Just another progress update :slight_smile:

Junk filter system updates

Yes, I’m still hard at work on this, as the definition of what I’m looking for keeps changing out from under me as I learn. Hopefully, that’s coming to an end. The algorithms themselves are quite tight by this point, I just keep flip-flopping on which ones I actually need and how to apply them :stuck_out_tongue:

The 2x, 4x, and 8x filters are also making their way into the system. It turns out that the 50th anniversary edition of Dark Side of the Moon is 24/96k in a 24/192k package, hence it gets the 2x cut. I do not have content at rates that would justify 4x or 8x, however… the biggest things in my library are 24/192k.

It also turns out that I’ve been approaching the spur issue wrong, as it’s incorrect to cut content to kill a spur. I only have three spurs in my library that deserve to be cut:

  • Nirvana’s Nevermind remaster at 24/192 has a nasty tone around 77k that takes the 50k filter
  • Sir Simon Rattle & the Berliner Philharmoniker performing Mahler’s 6th has a similarly bad spur at 48k that gets the 40k filter
  • Amnesty (I) by Crystal Castles has a tone at 44.1k; the album’s content is limited to 22.05k but it’s otherwise real hi-res and takes the 30k cut

Everything else’s tones cut through the content.

Adding the meter

Yes, it’s happening… the next version will have a Meter mode that shows the live meter and spectrum. Highly adjustable. The link will be next to the LIVE mode button.

2 Likes

We are long overdue, but it’s UPDATE TIME!

Up front: sorry, but this is not v1.16.0 with the spectrum and fixed junk filter auto-pilot; this is v1.15.3, and only accommodates HQP Embedded v6.1.0 with the two new 4-bit modulators and fixed backup zip bug.

I’ve actually hidden the meter page, junk filter suggestions, and auto-pilot from this release because the junk filter algorithm, as currently implemented, is worse than before (I don’t do my testing with the main container, I use separate scripts). With any luck, I’m only steps away from having it working, but I didn’t want to delay this update any more than I already had.

The spectrum sort of works, but the time axis isn’t accurate and it’s generally a visual mess. Far from ready for anyone’s eyes.

Another reason that it took me so long to release such a small update is that HQP 6.1.0 caught me in the middle of a massive, multi-day refactor after I tightened my gates up. Very few of the changes are user-facing… if you run into an error message, it should be clearer and more coherent now, as I restructured the internal error handling system, for example. Most of it just has to do with code correctness and legibility: principles like DRY (don’t repeat yourself), SRP (single responsibility principle), and test writing.

I’ve said before, I think, that around 50% or more of the time I spend working on HQPTuner is purely internal with no direct user-facing consequences. This ensures that the codebase remains legible and workable as its complexity grows. I take the maintainability of the codebase and testing very seriously.

Anyway, enjoy the new modulators! Those of you who can/do run DSD1024, anyway.

Now, back to this pesky junk filter problem…

2 Likes

I just wanted to drop in to say that, in spite of having my plate full with planned features that I have yet to complete…

  • Windows executables (still don’t have my Windows PC set back up :sob:)
  • Setup wizard
  • Filter playground
  • Junk filter recommendations and auto-pilot
  • A proper website (it’s in the works)
  • Meter and spectrum page

…I’ve been giving a lot of thought to what HQPTuner v2 is going to be like.

The one and only problem with HQPlayer, as I see it, is that there is no onramp for people to pick up speed before merging. That is, users are left to piece the “big picture” together for themselves.

Although it is one of my primary goals, HQPTuner currently fails to solve that problem.

My app is awesome if you already understand how to use HQPlayer. I am not afraid to say that it’s a big enhancement over the default UI. I’ve nailed the details, such as the filter narrowing system. And yes, I think HQPTuner is better organized… but the big picture is still missing. In other words, it lacks cohesion.

And that will be the primary goal of v2: rebuilding the UI around the big picture.

The gist is that everything will be focused around the signal chain, which will become interactive rather than just for display. Live settings like output mode, filters, dithers will always be on the main page, and deeper tweaks/settings will be found in “drawers”. This should, among other things, eliminate the need for Live mode while maintaining all of its features.

I’m also looking into how better to use the bottom Apply bar, such as pinning volume settings in the bottom left and moving the A/B switcher to the middle.

Every setting will be preserved; nothing’s going away. But the entire visual approach—the card system, the tabs, the cluttered top bar—needs re-imagined.

To that end, if anyone has any ideas or requests regarding how the next version looks and works, please do not be shy!

EDIT: yes, I am specifically building it to be tablet friendly :slight_smile: portrait and landscape modes!

2 Likes

Ok, I can’t help but show off one of my prototypes :smiley:

Note that this is highly imperfect (there’s literally still filler text below the Output options)… this is just an AI-generated, non-functional demo only for nailing down the aesthetic approach.

This is just an example of what v2 might look like.
EDIT: Image updated with a cleaner, better version

It’s a bit crammed in this view, but this is just to demonstrate that it fits entirely on an 11" iPad in landscape mode. Also, if the matrix engine is off or volume control is fixed, those go away to give you more space as well, and the Setting Switcher bottom bar is also optional.

The major features are all present on the main page: matrix profile, filters, shaper, output mode and rate. I’ve also managed to fit the EQ profile graph when applicable.

Filter narrowing features are available on demand from the Filter Narrowing button as a popover. Easy mode becomes the Filter Presets popover with far more compact tiles.

Presets have been renamed “Stations” while Live snapshots stay “Snapshots”.

Stations (presets) and snapshots are integrated into the same dropdown; snapshots can be saved to multiple stations upon creation and loaded with the station in a single click.

Instead of a setup tutorial/wizard, I’ve gone for the approach of “builder” buttons: a guided approach to creating matrix profiles, stations (presets), and snapshots.

All of the settings that you’re not seeing are accessed by clicking on the relevant portion of the signal chain on the left, or in settings via the gear icon in the upper right. The spectrum and meters would be accessed by clicking “Source”; output device and rate would be accessed via “Output”.

You’d be able to set whether settings stage or apply immediately; staging brings up another small popover in the bottom right with the option to apply or discard.

Once again, please do not be shy about feedback and suggestions!

2 Likes

Could this work with HQPlayer Desktop or Even V5

This would also make a killer plugin for LMS

Unfortunately, it will not work with Desktop, as it relies on the HTTP server offered by Embedded. I wish it did… I’d have a way bigger user base if that were the case!

It does work with Embedded v5 though! I know of at least one user who has it running on v5.

Is HQPTuner compatible with HQPlayer Embedded bootable 6.1?

Yes! …sort of.

I assume by “bootable” you mean HQPlayer OS.

It’s impossible to install anything else on the system running HQPlayer OS as I recall, so they can’t live on the same machine together, but you can absolutely run HQPTuner on a separate box. As long as they share a LAN, you shouldn’t have issues.

The one snag is that HQPTuner currently only runs on Linux through Docker, so if you don’t have any Linux boxes, the answer is temporarily “no”. Temporarily because as I am working on building Windows executables that should be dropping in the next couple of weeks (sorry this is taking so long y’all :sob:).

If you’re a macOS user, things are a little further out, as signing certificates are more expensive ($99/yr) and I’ve only received $20 in donations to that end.

In the meantime, if you have a Raspberry Pi or something similar, that would work.

Slightly confused.

I am running HQPlayer 6 Embedded on my Linux server and have installed HQPTuner on the same box and it seems to be running fine. Is this not meant to work?

One small fly in the ointment post installation. My UPNP was configured as auto in the hqplayerd config file. After installing Docker and HQPTuner I found I that HQPlayer was no longer detected as a UPNP renderer. Then I discovered that the network binding was appearing to default to docker0 instead of my network adapter. So changed the setting from auto to the network adapter id and the renderer appeared and everything is working.