[vergesslich2]

Hi! Sorry. I have absolutely no clue. What kinds of artefacts can be heard, which cannot?

Several Problems. I didn't go the standard route in APO I guess. I used RBJ biquad filters to get better sound out of my headphones. I remembered I could use biquad filters, because ... they have been there for a long time. But I actually just wanted to see round things in the graph in APO and thought that it also had less computations than the graphical straight-line-tool-EQ (it uses fftwf_execute, so FFT in general, correct?).

I thought, RBJ Biquad would give not as much latency when making music and using APO. But maybe I could mix filter types, each corner of the spectrum gets the adequate filters? Like... Biquads for low frequencies, FIR / fftwf_execute for high frequencies? There the filter kernel can be smaller, right? So... does anyone mix them or am I overthinking it already?

Another thing is the ringing types. Do you hear such things? I heard it when I tried IIR vs FIR. But I'm not sure what I heard (FIR just sounded a bit dryer), and it's been long ago, and now I have this 560s headphone and am getting used to it.

My (?) idea was (back then) that pre-ringing plus post-ringing sounds dry, because it's symmetric. And post-ringing-only in my mind would sound wet, because it's post, like a response, resembling a reverb-system, a bit. I guess that could be very wrong. I'm here to stop thinking wrong things. ***

Another thing is, now that I have chosen RBJ biquads, I'm bound to the sample rate, I believe. APO with the biquads uses the sample rate that has been requested from the device, by the application, and that has been Firefox, requesting 44.1 Khz. The biquad slopes should sound kind of the same with 48 and 96 Khz, but I'm not sure if they do with low Q values. Do I need math, or can I go around math? The exp/log stuff is hard for me, everything in math is.

What I know is that slopes near Nyquist will sound different from 48 Khz to 96 Khz for example, because the filter - in layman's terms - "needs to go to zero amplitude response" at Nyquist, and it has less samples to do so when there's only 48 Khz instead of the 96 Khz. So with 48 Khz the slope is steeper, right? And my problem might be a bit complicated, sorry: I placed a negative-gain peaking Biquad before Nyquist, because I wanted to save computations. If I had a shelving filter and wanted my preset to have "air" at 19 Khz, to fight the roll off, then I would need two filters instead of one filter. One that shelves the harshness out (headphone is too harsh), and the second would make it go up again. Would this be the more correct way for getting the same result with 96 Khz, or do shelving filters near Nyquist change their slopes just the same?

Sorry for so much text. I'm interested, but unqualified. Also I'm only at page 24 in the Floyd Toole book, and having no DSP background.

If I can leave anything out, because using it cannot be perceived, then I will do that. I have attached the APO config.txt in question.

Thanks for reading.

Edit: I found a discussion about similar problems. Looks like it could be cheaper for me to just make sure to always use 44.1 Khz, because I'm used to Firefox and that seems to rely on it. https://sourceforge.net/p/equalizerapo/discussion/general/thread/e2fd040bcc/

Edit2: I might have progress, but not sure. With Nigel Redmons biquad designer (unfortunately its shelf-filters provide no Q, and APO provides no way to enter the sample rate) it appears, I could maybe just cascade the same filter for getting a similar (?) slope at a doubled sample rate. And for 48 vs 44.1 I also could cascade and make the "second pole" or second filter have a lower Q to get only slightly increased steepness? I didn't understand if the links on the equalizerapo discussion page all lead to how to transform something to FIR, or if some link shows how to just cascade biquad filters. I'm lost. But I tried.

***
My memory was wrong. Better conceptualized and correct information is here:
https://producerhive.com/ask-the-hive/linear-phase-eq-vs-minimum/
https://www.adventures-in-audio.com...se-eq-on-transient-signals-such-as-snare-drum
https://troll-audio.com/articles/linear-and-minimum-phase/
https://audiophilestyle.com/forums/...porware/page/699/?tab=comments#comment-983918

 

[ADUHF]

Some of your questions above are above my paygrade, vergesslich2.

I loaded the parametric filters from the above config file that you're using into Equalizer APO's Configuration Editor though, and got this...

The plot may look a bit different on your system in the higher treble, because I'm using a 48 kHz sample rate for better compatibility with YouTube and some of my other video hardware, rather than 44.1 kHz.

I don't know if this applies across the board to all Equalizer APO projects, but when I try to use some of the Parametric filters (such as the Peak filters) in their Configuration Editor, I do notice the functions begin to distort as you approach the Nyquist frequency, which is 1/2 of the sample rate. So on my system that would be at around 24 kHz.

If you simply want to shelve the treble down though, then there are couple different ways of doing that, which should eliminate the rise in the treble near the Nyquist frequency.

You can use either a Low Shelf filter, combined with a negative Preamp...

Example:
Filter: ON LSC Fc 750 Hz Gain 10 dB Q 0.5
Preamp: -10 dB

Or you can use a High Shelf filter with a negative Gain value (without the need for a Preamp)...

Example:
Filter: ON HSC Fc 750 Hz Gain -10 dB Q 0.5

I think either one of these should potentially work ok. And they will both produce an EQ curve that looks like this, with no rise back up in the high frequency range.

Whether that will help you at all, I don't really know.

I have played around a bit with some of the various Parametric EQ filters in Equalizer APO's Configuration Editor. But I mostly use it for doing (and combining) various plots with the Variable Graphic EQ filters.

 

[vergesslich2]

Thanks!

Babble:

My math and DSP background is weak. The questions I asked are the ones where I get into a confused state and say "f..k". I had the idea to ask the people in the DSP forum on KVR. Their understanding is ... well some over there begin laying out math quite fast. And people with DSP knowledge have impressive understanding of ... well.. DSP.

Problem for me: In school, when we had to do limes, Euler, derivations with sin(), and things that have some complexity to it, I sat at home before my computer, later got F marks, started to hate math. Now it haunts me when I look at DSP topics, because they look the same, or are the same. I look at it and my inner self immediately says "*** that". Also, I still hate the fact that math exists as line based math like in Basic or C, and as LaTeX, so to say.


Important part:

So, the "down-peaking" filter distorts near Nyquist? This is good to know. I mean, I don't want distortion. If I go on with it, I'll use shelf-filters.

I had used the negative-gain peaking filter on purpose, because I actually wanted the frequency response to go up right somewhere before Nyquist, because I had the impression that the HD 560s also had some early rolled off highs, not enough extension.

And while doing this, I wanted to spare off one filter (shelf-down, shelf-up <- this one). So I decided for the peaking filter.


Babble:

Now it's like this: I gave up on the HD 560s. At first I tried another DAC, and that indeed made Firefox use 48000 Hz out of the box (the other DAC made it use 44100), and I already got the problem that I had worried about.

At first, I had somehow these two ideas, I just try to explain again:

1) Accept the (big) latency. It's there even when the config.txt uses only biquad filters. I didn't understand that part at all. It was not 10ms latency, it was 1 second latency (Firefox -> APO -> Roland MV-1). So I could just use the graphical EQ tool with straight lines.

2) Use code that plots an EQ (Nigel Redmon has an open source biquad designer), or FFT the impulse response (maybe this would be safer? But I trust Mr. Redmon quite a lot, and there's very concrete math for the plotting, one just needs to use double precision AFAIR because of some log() part.), to get the frequency response and then generate the 48 and 96 KHz parts of the config.txt.
With the frequency response, I thought, I could just use Newton's method to find new Fc and Q values. And if needed, for more steepness, add more filter sections or something like this.

But the thing is: Some people surely know how to handle samplerate changes in their EQ code. They know DSP, I don't. I don't trust my few known DSP facts, and I want no two weeks adventure trying to make the 560s sound nicer. And I did not like Sonarworks at first glance (needs account). I just wanted to have determinism in a user interface (APO), yet again. So: I gave up and ordered the HD 660s! (Edit: Reality: ) I hope the parcel will arrive today.

Thank you sincerely for taking the time.
Kind regards

 

[ADUHF]

Parametric EQs take some practice to use well, I think. And I'm still not sure I know the best ways to go about it. The fewer and more well-targeted they are though, the better I suspect. Simply because its less hassle to keep track of. I've probably only scratched a bit of the surface though for how these tools could be used.

The Nyquist thingy is also somewhat problematic when it comes to making any sort of adjustments in the higher frequencies. At least it is in Equalizer APO's stock Configuration Editor. And for that, and also other reasons, I'll often just throw my hands up in the air, and go back to using a graphic EQ approach instead. (Which I happen to think is also perfectly ok.)

Some like the other front ends though, like Peace. Which I've never tried, and don't have much interest in.

The method that I'd probably use to build an EQ curve with parametric functions would most likely all be done graphically in the Configuration Editor though. Rather than using mathematics. Because that is just the way my brain is oriented.

I'd probably start by roughing or sketching out my EQ curve with some other tools that I'm more comfortable with, like one or more variable graphic EQ filters. And once I'm more or less satisfied with the sound I've got with that approach, then I'd try to figure out a way to convert that overall EQ curve into just a relative few, well-targeted parametric functions... Easier said than done, right?

The stacking interface, and Analysis Panel in the Configuration Editor make that somewhat easier to do graphically though. (And maybe at some point I'll try to do a video or tutorial to show just how it can be done in there.)

One useful tip to remember on this is that most of the parametric filters can take either negative or positive values. So its very easy to flip their polarity by simply switching the gain setting on the filter from positive to negative.

And the same can also be done for a graphic EQ curve using the inversion tool in the toolbar in the lower righthand corner of each graphic EQ's window...

A more close up view of the tools...

The "invert response" tool is the middle tool in this set. And it will mirror or flip the entire graphic EQ curve upside down, making all the positive values on the graph negative, and vice versa.

So... if you start with a graphic EQ curve, you could just potentially flip that curve upside down, and then use various parametric filters to smooth that curve out into a flat line in the Analysis Panel. Then if you turn off the inverted graphic EQ filter that you were using as a guide, what you should be left with is a group of parametric filters that should pretty closely match the shape of your original graphic EQ curve before you inverted it!

That might be one way to approach it anyway.

There are some others though that are a bit more involved, where you'd start with an approximation of your headphone's response curve flipped upside down, and then conform that shape to another target curve. And both of those curves could potentially be plotted as their own subsets of parametric filters. This could potentially get quite a bit more complicated though. But the two curves (rough inverse of headphone's FR, and target curve) could both be plotted separately with parametric filters, in exactly the same manner as I described above for matching a single graphic EQ curve. And then they could be joined together by combining both sets of parametric filters into one single config file.

This is not something I've actually tried though yet. So I'm not really sure how practical it would be.

 

[ADUHF]

Are you going to be hanging onto the HD 560s btw? Or will it be replaced by the HD 660s?

By most accounts the HD 560s isn't too bad for a pair 0f $200 headphones. And I've been sort of thinkin about trying a pair myself, since I don't currently have any open-backs, or Sennheisers.

 

[vergesslich2]

Are you going to be hanging onto the HD 560s btw? Or will it be replaced by the HD 660s?

By most accounts the HD 560s isn't too bad for a pair 0f $200 headphones. And I've been sort of thinkin about trying a pair myself, since I don't currently have any open-backs, or Sennheisers.

It's complicated.

I have returned the 660s. It was the B stock sale from Sennheiser. Its frequency response was correct, and I wouldn't have had any idea what I could improve with an EQ, so I didn't even try. But the sound, without EQ, was not what I had expected.

I try my best explaining it. The bass** was like a cloud. Some highs, when I noticed them, sounded like added on to the mix, maybe being not smooth. Details were like "acoustic impressions" that "went through the cloud". Loudness of parts (depending on frequency? original loudness? I don't know what was going on) of the music was like "compressed upwards", which made it "sound constant", and gave every track I tried the same sounding. Sound stage, compared to 560s, was narrow and in my head.

I'm no 560s fan, because my 560s sounds so bright, but it had none of the problems the 660s had. My 560s are very bright, that is the only thing they do wrong for me. I keep them for the digital piano, and maybe for using it with Equalizer APO. Without EQ I cannot use them, too bright.

I sold my HD 600 (which I found just perfect) in 2019, so I don't know if any of my assessments of the HD 660s is correct. Maybe I wanted the 660s to have the sound stage and features of the 560s. With EQ the 560s keeps its sound stage and some of its dry presentation or linearity, so I just expected too much from the 660s, I guess.

Edit: The 560s bass** can sound like "pudding" too with my digital piano, but I didn't find it worrying.

Regarding the problems:
Pudding sound:
560s bass** = more the lower bass than the upper bass
Cloud sound:
660s bass** = lower and upper bass, maybe even the lower mids