HQ Player

[ray-dude]

It is a HQPlayer Desktop option (on the tool bar at the top of the window), but I.  When I exported setting from HQP Desktop, I didn't see anything obvious in the XML for inversion (HQPe settings XML doesn't have anything obvious either).  It may be a playback software option rather than a HQP engine option?

[ray-dude]

24 minutes ago, Miska said:

 

It is option in HQPlayer engine, goes together with volume control. But it is not related to configuration.

 

 

Thank you @Miska.  

 

I've been doing a lot of reading this week trying to grok the various settings and configuration options and have some questions I haven't been able to find answers to.  

 

• In the configuration XML, out of curiosity what do "fft_size" and "dualwire" impact?  
• To completely disable volume control in HQPe, I'm setting volume_max, volume_limit, and volume_min to "0".  Is this the right way to do so?

 

More generally to the broader group, I'm using a Chord DAVE for my DAC (PCM biased), and was hoping to tap into the experience of other Chord owners.

• Is there a consensus best filter for SDM to PCM conversion for Chord DACs? (running on i7 NUC)  
• For Chord owners out there, do you find that you prefer the native DSD pathway in the DAC (presumably with DSD to DSD upsampling), or better to convert to PCM and pass 16fs upsampled PCM to the DAC?
• DAVE supports 32 bit PCM, so better to feed it 32 bit or 24 bit? (not sure if it prefers 24 bit internally)

As I come to a preferred configuration with Chord DACs, I will be sure to write up my learnings and findings (quite the journey to navigate 650+ pages of contributions!)

 

Thank you to everyone that has been sharing their HQP journey and insights here (and to @Miska for an incredible piece of software).  Fantastic resource, and I've learned a heck of a lot.

 

[ray-dude]

Since you're a Roon user, as a workaround perhaps use Roon DSP to do the inversion? (under DSP settings -> Speaker Setup)

[ray-dude]

@miska I just saw the release notes for the new HQPlayer Embedded.  Do you have information you can share on the new sinc-S filter? 

 

Looking forward to trying it when it makes its way to HQPlayer Desktop!

 

[ray-dude]

On 5/28/2020 at 3:07 AM, Miska said:

Note about the new sinc-L filter (in HQPlayer Desktop 4.6.0); to use this one especially to higher DSD rates you'll likely need a GPU. To DSD256 it requires about 8.3 GB of GPU RAM, so my RTX2080 runs out of memory and thus doesn't play. However, on RTX2080Ti plays fine to DSD256 with a bit over 10% GPU load. For DSD512 (not tested) you'll likely need TITAN RTX with it's 24 GB of RAM should be enough for the ~16 GB needed.

 

Memory bandwidth is biggest bottleneck for massive filter on CPU because it doesn't fit in the cache anymore.

 

At DSD256 number of taps is about 32M and at DSD512 64M. DSD128 with 16M taps works on GPUs with less RAM.

 

With 16x ratio to 705.6/768k PCM it is not so bad with 2M taps.

 

P.S. The filter is totally different kind of design from sinc-S/sinc-M although named similarly.

 

@Miska I'm looking forward to trying sinc-L this weekend!  Could you speak more to what you were trying to achieve with this filter vs sinc-S/sinc-M and filter performance? (appreciate the insights you shared on the sinc-S filter a couple weeks ago)

 

Looking forward to hearing this one!

 

[ray-dude]

It can be hard to get started on HQP, just because all the choices can seem so overwhelming.  Once you have a working setup that you like, much easier to tweak up different filters and settings to tune to your ear.

 

As Rajiv and others have mentioned, with TT2 you want to target 16fs (768kHz for PCM, 44.1*256 for DSD).  sinc-M + LNS15 will give you a sound profile closest to Hugo mScaler.  Setting Vol Min and Vol Max to -3dB will force all volume control to be with your TT2, and give headroom for the upsampling so you don't get clipping.

 

A lot of the rest of the parameters are dependent on your streamer and preference.

 

There are MANY other Filter and Dither combinations to try.  Once you get sinc-M going, try sinc-L and sinc-S (I find sinc-L to be quite nice)

 

Once you start finding a preference, it gets easier because you can start exploring other filters within a family (MP vs LP filters, etc).  I happen to be a hard core LP guy, but others really enjoy what MP brings to the party.  

 

I've found that the more your ear gets accustomed to what HQP is doing, it becomes a lot easier to dial it in to your taste and your equipment (alas, there is no optimum configuration, which is why there are so many options!)

[ray-dude]

Depending on the filter you are using, yes this is normal.  With the sinc-M filter, believe it or not the sound you are hearing is processed from the recording in the future and the recording in the past, so HQP needs to accumulate data for a couple seconds before it can calculate what you should be hearing now.  For the gory details, see:

 

https://en.wikipedia.org/wiki/Whittaker–Shannon_interpolation_formula

 

As you use different filters, they will require a different window of time for their processing, so you'll have different delays.

 

 

[ray-dude]

53 minutes ago, Miska said:

 

Dither / noise-shaper is way to limit output word length (re-quantize) in a way that doesn't produce distortion.

 

For Chord DACs, stick to 705.6/768k output with LNS15 noise-shaper. Alternatively you can also use for example TPDF dither. If you use USB input, leave DAC Bits to Default. If you use dual-S/PDIF with adapter interface, set DAC Bits to 24.

 

 

I can only speak to Chord DACs, but in addition to the above, DAC bits and buffer both serve as final "tone" controls for me (I use USB from Taiko Extreme to my DAVE).  Sound is most incisive sounding at 32, as I step down I can even out the mid range presentation (I'm usually around 29 or 30). Buffer time is less impactful for me, but does impact dynamics for me (this may be more system specific than DAC specific). Higher buffer is slower (but fuller), with lower buffer more dynamic (but perhaps thinner).  I'm usually 20 or 50ms.

 

These really are final tweaks for me.  Get your preferred filters and noise shapers, then use these settings for the very final tweaks.  In general, I've found that the more I improve my digital chain (power, etc), the more I'm able to push incisiveness and dynamics and still have things sound natural and relaxed.

 

YMMV of course...this is subtle stuff, and very system specific.

 

 

[ray-dude]

 

 

Here are my current settings.  sinc-M and LNS15 gets you mScaler-like, and adjusting buffer time and DAC bits gets you that final tuning to your ear.

 

I find the sinc-L quite nice as well (more for some albums than others), but I keep coming back to sinc-M

 

 

 

 

[ray-dude]

The modulator section only applies if you're upsampling DSD content.  Tidal content is PCM, so it goes down the PCM Defaults path.  If you happen to play any DSD content, it will go down the SDM defaults path, and use that modulator.  Choice of modulator is dependent on the power of your computer (some are VERY compute heavy) and what sounds good to you.  In your case, 99% of the time it won't matter.

 

If you set Vol Min and Vol Max both to -3dB, it will fix the output volume, and you will control volume on your DAC.  As background HQP can also control volume, and these settings set the min and max volume settings for HQP.  If you're controlling volume with your DAC (which is what I do), setting both to -3dB basically sets HQP at max volume output.

 

 

 

 

[ray-dude]

A very material upgrade for NAA users! (PCM guy as well, in case that makes a difference) @Miska any hints on what types of NAA optimizations could have this kind of impact?  If there is any more that can be done in this same direction, yes please!

 

[ray-dude]

4 minutes ago, Miska said:

It has now one more (very thin) layer of buffer and one extra thread for dealing with sending the data over network, this allows CPU staying a bit more busy doing DSP work (less likely to have drop-outs). One could say that the NAA output case is now one more step "pipelined" and asynchronous.

 

This is fantastic!  Any way to tweak buffer size to see where an optimum point might be?  This is definitely a needle mover, and most appreciated!

[ray-dude]

1 hour ago, Miska said:

Twitter and Facebook. So it is opt-in in this way...

 

 

I use https://visualping.io to monitor the website for changes, and email me when it does.

[ray-dude]

1 hour ago, Miska said:

Quick update with 4.9.2, I hope I finally found the root cause and nailed the bug. I also made some fixes to the manual.

 

 

Looking forward to listening to this later today!  Most of the NAA magic of 4.9.0 was lost with 4.9.1, so very hopeful to hear it again!

[ray-dude]

11 minutes ago, Miska said:

I don't know where that magic comes from. Just an occasional problem starting playback that I've been fixing... (because it was trying to send a wrong message at wrong time)

 

Well the magic is awesome!  If helpful, happy to help figure out what is happening on my system so we can take advantage of this very welcome coincidence (I've heard similar reactions from several other people, but it may be very system dependent)

 

 

[ray-dude]

1 hour ago, k6davis said:

Here's how to get HQP OS/HQP Embedded running with Roon. This will require at least 2 computers: 1 for Roon and another for HQP Embedded. You can try it out without making any changes to your current server.

 

Great recipe!

 

To pile on here, if you are able to try this (2 computers etc), you most definitely should.  Booting off a USB stick makes it a very experiment to do (then undo).  Delightful how many problems and annoyances simply go away when you have a stripped down end point that is optimized for one thing and one thing only.  

[ray-dude]

 

When not crashing NAA or glitching during playback, @Nenon's experience is the same as mine. 4.9.0 was a significant step up in SQ with a dual CPU system, HQPD running on one CPU, NAA running on the other (I believe @Nenon is running a similar config)  4.9.2 (with the fix) gave up some of that, but not all of it

 

I suspect it is not the bug but how data is being passed between processes?  If there is a way to understand that and optimize it, that would be a huge step forward (the improvement with 4.9.0 was decidedly non trivial)

 

If you're not hearing it in your test setup @Miska I'm sure this is a frustrating discussion.  This may be a very system dependent phenomenon.