Which HQP Filter are you using? [2015-2023]

That’s a good starting point. Please see the config help page for details and some suggestions regarding filters. But don’t get too much bound to those suggestions.

RPi4 doesn’t quite have enough processing power for doing proper DSD upsampling…

So if I wanted to play with DSD, would an investment in Up squared be a solution. Or do I need to go the Desktop version and let my i5 NUC do the work?

It depends on what modulator and filters you wanted. I would start with what you have to get a feel for what you might (or might not) need.

I thought I was listening to R-2R?
44.1 → gauss-long + ASDM5ECv2 → DSD1024
Felt quite relaxing and balanced. DAC was Denafrips Ares II.

5 Likes

I assume i9-12900k? Is this with CUDA?

Yes. It’s 12900K. CPU OC 51x, ring OC 44x.

Could do it without CUDA.

Using NAA would help a lot. USB thru ALSA would increase the utilization of the two primary P-cores).

1 Like

Was a lot of techs and specs here latetly let’s talk about music for a change! :slight_smile:

Caveat: I have R-2R DAC and I listen via floor standing speakers, i.e. there’s a room factor in what I hear (beside my ears, brain and general listening habbits)

In last few days I have been a bit puzzled with my listening experience and impressions. In regular “Your Daily Mixes” section ROON has offered me a “Radiohead” playlist. Excited, since I have not visited Radiohead and Co for a veeeeeeeery long while, I plunged into listening.

Annoying! This first reaction of mine caught me by surprise. Noisy, messy with poor channel separation, with voice drowned into the wall of sound. To be specific, here are few things that “kicked” me at the start of the mix (streaming from Qobuz):

  • Radiohead - Lewis (Mistreated)
  • Interpol - Everything is Wrong
  • Arcade Fire - Neighborhood #1 (Tunnels)

My latest default HQPlayer setings that digest a large volume on genres are 7ECv2 SDM@256 with:

1x poly-sinc-short-mp
Nx poly-sinc-gauss-hires-mp

First things first - I started to play with the filters. After multiple tries, including the “ext2” and “gauss” group of filters in SDM I ended up with 1x poly-sinc-gauss-hires-mp

I switched to PCM. IIR LNS15@1,5/1,4Mhz was producing smth less fatiguing, less irritating and more coherent.

I wonder what was it ?

Why “alternative” listening experience would be so much different technically to “prog” which is lately one of my main listening streams, and that is excelent with short-mp I largely suspect the recordings them selfs are the culprits, since not all the Radiohead is annoying like that. But equally this way of recoding seems to be intentional (to be frank, the worst was Arcade Fire)

I have looked at some locally saved files, et voila, no wonder I was knocked back earlier. Look at the waveform ! With a “whopping” dynamic range of 3. The source is “Hi-Rez” 44,1/24, btw…


So, I concluded I was “kicked” by that “loudness war” production. But equally - not all “loud” recordings with low dynamic range are that “bad”, the best example for this is Beck “Morning Phase”, “Colors” or “Sea Change”. Eh, I wonder…

What’s your HQP sweetspot for “Alternative” ?

What’s your HQP experience with “loudness war” recordings?

@IgorSki there just is no perfect filter for all occasions as you know but, I think that for most things pop/rock, poly-sinc short mp is hard to beat. Also, PCM 768 with NS5 betters DSD for kicking out the jams. All just my opinion, of course.

1 Like

Ever since gauss-hires came out I’ve been using the following combination with ASDM7ECV2 / DSD256

Gauss-long (1x)
Gauss-hires-lp (Nx)

The combo is the most versatile I’ve found, it evens makes some poorly mastered recordings that sounded like utter crap on some of my previous favorite filters (such as ext3 & xtr-short-mp) sound divine.

If I do switch to anything else it’s not for long and I always come back to this combo.

4 Likes

Yup those are good for some music, I listen to them here and there. Question though, why wouldn’t you listen to gauss-long at nx?

Hey, thanks for a tip, I’ve not really thought of stepping down from 1,4/1,5Mhz before, somehow. Althought I’m not a frequent PCM goer, surprisingly to me, I liked 768 in “HA MAY” ! Except, I was switching back and fourth between NS5 and LNS15… do I hear the difference? :slight_smile:

Yeah, I’ve got mine “HQP swiss knife” combo as well. But this time (I’m talking about my post above), it did not work. After my panic attack and some listening it all boils down to these two albums, really. Try them, interesting to know how you accept them in your system, with your listening habbits + HQP set up. In my case, I guess my ear-brain system was just not ready, and I struggled :crazy_face: :nauseated_face: :hot_face: :cold_face:

1 Like

I can tell you Funeral sounds fantastic with the above settings in my system, it’s in my regular rotation (as does Radiohead / Thom Yorke in general it’s just been a minute since I listened to My Iron Lung…I’ll have to get back to you on that one)

1 Like

Thanks !

I than strongly suspect it is just my listening habbits. “Funeral” is a sound scandal for my brain :slight_smile: As for Radiohead example above, not all compositions kick me in my back. In fact I hear “My Iron Lung” as more clean recording comparing to “Funeral”, judging by my ability to separate instruments and follow its’ lines.

Interesting… I will continue my quest :slight_smile:

Over this weekend, thanks to Roon daily mixes, I have discovered tons of Italian progressive form 70s, wow!

1 Like

Looking back at your post, something seems off with your copy of In Rainbows. Even for a loudness war album the lowest dynamic range it should have is a 5 (high quality 24 bit would be a 7). It would go up from there most likely indicating the quality of the rip since they are vinyl after that. If your dynamic range is a 3, it sounds like you have a transcode (which is likely what you are hearing).

PM me so we can keep this thread clean.

https://dr.loudness-war.info

1 Like

Thank you, sure I’ll PM you! Indeed we are gradually drifting off-topic.

Nonetheless for “public” eye and to share experiences the question is still valid

  • What’s your best HQP filters for “loud” and heavily compressed recordings?

PS: btw, on your remark above - “loundess-war” database is “crest method” DR measurement, in ROON is R128 “loudness” standard. Different algorithms, can’t compare directly. The simpliest and best ever explanation on differences I saw here Dynamic Range - JRiverWiki

1 Like

Just IMO, subjective opinion.

If you have a really dense (or compressed/clipped/loudness war) recording, it is even better to stick to higher rate DSD output. At least DSD256 or even better DSD512, using the EC modulators.

PCM conversion, even at 1.5M, can become congested and lack “depth and air”.

If it’s 44.1/48k source, use of an apodizing filter is typically must. Also good to stick to some shorter filter, such as poly-sinc-short. And possibly minimum-phase.

Another approach is to use longer one, but one that is especially transient optimized, such as poly-sinc-gauss family.

I’m nowadays very happy with
1x=poly-sinc-gauss-long
Nx=poly-sinc-gauss-hires-lp
So much that I have made this the new default in latest release.

12 Likes

For anyone using the Convolution Engine, or @jussi_laako does enabling a convolution algorithm add any significant additional CPU or is it negligible?

(Currently upsampling to DSD256 using an M1 Mini for reference)

Depends if you’re doing convolution with DSD files?

For PCM files, it’s not significant.

Can be significant for DSD files, especially multichannel DSD input and DSD output.

Just give it a try on your M1?

I normally would but the specific filters I’m thinking of adding are $200 (LCD-5 FIR Filters from Mitch Barnett). More details in the link below, I’d like to be relatively sure before placing the order though (scroll to the bottom for details on the filters)

What you just said confirms my understanding though. I’ll only be applying the convolution to PCM (not any native DSD which would need to convert to PCM first at 1/16 DSD rate, which could be CPU intensive). No multichannel either.

He offers the filter up to 384k and will split the L&R mono as well.

Have you also tried oratory1990’s LCD-5 EQ?

No need to use FIR convolution for that. Just use HQP’s built-in parametric EQ.