New Filter request(s)

[acg]

I will compare tomorrow with 1.186-d and original AP that I was running before my holiday, but 1.186-i does sound very nice.  

Hi Peter,

I am going to need to spend some more time listening to the Custom Filter to get a good handle on it.  It definitely sounds "warmer" and I do enjoy the resolution of steel strings that it has enhanced but I just wonder if some things sounds a little bit "the same" i.e. not enought variation betwwen the same instrument on different albums.  Not sure at the stage.  All was going well until I put of a couple of "warmer" sounding albums and I was not so convinced.

Anyway, I will listen for another week or so and then switch back to standard AP and report back.

Cheers,

Anthony

I have lots of hours up with the new Custom Filter in 1.186i with my NOS1 (not the a version) and I was really enjoying the music today so I thought it would be a good time to check back to standard Arc Predict.

Well, what to say, but Arc Predict sounds 'truncated', like the top end has been removed.  It is flatter sounding and a bit dull by comparison.  There is less jump and less dynamics and I guess that is to do with there seeming to be less HF information.

So wow Peter, I think that this filter is a very significant step forward and I can't wait to see what else you come up with.

Can you please tell me which sampling rates this filter works with or does not work with?  16/44.1 obviously is included, but are others like 24/192 or 24/88.2 etc. included in the supported Custom Filters? 

Cheers,

Anthony

[PeterSt]

Hi Anthony, thank you for that feedback.

16/44.1 obviously is included, but are others like 24/192 or 24/88.2 etc. included in the supported Custom Filters?

At the moment I would allow to set all parameters by yourself instead of through the presets, yes. But this not assumed (it will be chaos) at this moment it requires more : the selection of the native sampling rate (together with the output sampling rate as it exists at this moment). This is not all that difficult to make.

What will be more difficult for me is finding the merit of such filter (parameter) settings. I mean, I don't think there will be any logic in it for me, so I would not know what to look for (best settings hence best output signal). And I think that chances are very fair that using native Arc Prediction will just be the best for the higher native sampling rates (88.2+).

Notice that a digital filter has two purposes :
1. Reconstruction (really only applicable to 44.1/48);
2. Bringing down high frequency "noise" (same signal but images beyond Nyquist (sampling rate / 2)).

What Arc prediction will do regarding Hires is #2 only. Oh, it will work on #1 as well, but this is quite moot because not needed unless we think that frequencies between 22.05 and 44.1 (for 88.2 sampling rate) are very audibe and need reconstruction too.
So what's left for the Custom Filter is bringing down the HF "noise" more but to a degree which is rather moot (because Arc Prediction at 705.6 (768) output already has it down sufficiently to be harmful (beyond 352.8 (384)).

So it's almost the other way around : When the Custom Filter is engaged by standard (for say 705.6 and anticipating 44.1/48 playback) it should not engage the filtering itself when Hires is in order and leave Arc Prediction as it is.

All again will be different when Arc Prediction is shut off in the first place (see grayed checkbox in the Filter Designer) and now *only* the Custom Filter is operative. But here too, you won't be able to make it yourself because it needs to show the results of it (though this I could graph) and now I also don't feel much like making such a filter for you guys because I just don't see the reason (for the better) for it.

Hopefully this is not too vague (I did not try to be this time ).

Regards,
Peter

[charliemb]

So it's almost the other way around : When the Custom Filter is engaged by standard (for say 705.6 and anticipating 44.1/48 playback) it should not engage the filtering itself when Hires is in order and leave Arc Prediction as it is.

I think that this is true for 705K.  But don't forget us peons who can only go to 384K! 

[PeterSt]

for say 705.6

Charlie, the "say" was meant to be an example. But I also think that maybe you didn't get what I meant. So for 352.8 and the others it would count the same (don't engage that filtering). Tell me if it is not clear.
Peter

[charliemb]

Notice that a digital filter has two purposes :
1. Reconstruction (really only applicable to 44.1/48);
2. Bringing down high frequency "noise" (same signal but images beyond Nyquist (sampling rate / 2)).

What Arc prediction will do regarding Hires is #2 only. Oh, it will work on #1 as well, but this is quite moot because not needed unless we think that frequencies between 22.05 and 44.1 (for 88.2 sampling rate) are very audibe and need reconstruction too.
So what's left for the Custom Filter is bringing down the HF "noise" more but to a degree which is rather moot (because Arc Prediction at 705.6 (768) output already has it down sufficiently to be harmful (beyond 352.8 (384)).

So it's almost the other way around : When the Custom Filter is engaged by standard (for say 705.6 and anticipating 44.1/48 playback) it should not engage the filtering itself when Hires is in order and leave Arc Prediction as it is.

All again will be different when Arc Prediction is shut off in the first place (see grayed checkbox in the Filter Designer) and now *only* the Custom Filter is operative. But here too, you won't be able to make it yourself because it needs to show the results of it (though this I could graph) and now I also don't feel much like making such a filter for you guys because I just don't see the reason (for the better) for it.

Let me clarify what I understood because I think things are different at 352/384 than they are at 705/768 and I think that even hi res needs help.  I understood that you might make a custom filter that behaves like AP when the source is hi.res. and like Custom (how it is now) when sourced with 44/48. 

What I'm saying is that even with hiRes source material, at 352/384 my ear is much preferring Custom over AP.  Hard to argue with the ear.

As an aside, I now hear the buzzing / zooming / mosquito-ing of the typical upsamplers that you've talked about.  I also hear that AP never has it, nor do the interpolative filters in HQPlayer.  With Custom, I can *sometimes* hear a little hint of the buzzing.  I accept it as it comes with a bunch of other goodies.  But usually when I hear it with Custom, it is in the recording itself; and I presume this is from the EQ processors that audio engineers use for vocals.  No doubt they ring, some more than others.   And I can check with AP for those recordings and sure enough the ringing is there for those recordings.  It's not a clean ring when it is in the recording but the essence is surely there.

Anyway, I hear the trade off that you are making with these filters and can see that it is like walking a tight rope, picking the right balance between the two extremes.

I can say, it is amazing that Custom measures better than AP because AP sounds cleaner.  But sound is sound.

[PeterSt]

All right. By now I may think differently again and that even Hires should be filtered in the "non-AP way". So no worries.

Also : Because at this moment you can only use the 176.4 filter while using 352.8 this will ring unnecessarily.
Nothing wrong with your ears !

Peter

[charliemb]

An attenuation of no less than -6db has been recommended with the custom filter.  In my case I'm talking about the 176 filter, used mostly at 352 and also at 176.

For sure audible clicks are heard at 0db.  For a while I ran at -3db of attenuation because I'm running deep class A on my ams and due to heat I have limited headroom.   The purpose if this post is to confirm that -3db is not enough.  When I run at -3db I don't hear clicks, but it is possible to hear things that don't sound right.  Further attenuating to -6db solved that problem. 

It's still hard for me to believe that the intentional ringing is reaching with peaks of  200% which corresponds to -6db,  but it appears to be the case that 141% which corresponds to -3db is for real.   

I've not tried -4.5db and may never try it unless I actually need the headroom / volume.

[charliemb]

This is regarding the 176 Custom Filter, run usually at 352.

The purpose if this post is to confirm that -3db is not enough. 

Correction: The quoted text above (from me), while true sometimes, is not true most of the time.  It depends on the compression level of the recording.

I've been running three weeks now at the recommended -6db and xx has turned out to not be my favorite player in this time.  The sound became drier, even with the custom filter which had solved the dryness.  AI sometimes came to the rescue but takes too long to load to use every day. 

So yesterday, I came across a dry sounding situation that was only dry with xx and not with other players.  Hmmm.  This was on the highRes version of the Indigo Girls All That We Let In,  This recording I know has peaks that never go above 90%, and I was upsampling from 96 to 384.   So I changed the attenuator from -6db all the way to 0db.  The dryness was gone!!!   

The same is holding true for most recordings that I've tried since then.

And so it appears that at -6db there is less "ringing" going on, and less reconstruction of the higher harmonics, than at 0db.   

My standard setting right now is -1.5 db and if I hear anything funny I drop to -3 db or -4.5 as needed.  The more compressed the recording, the lower I must go.

[PeterSt]

Thank you for your effortless sharing Charlie !

Peter

[charliemb]

I've been trying many different settings for the new AI, including all of the dither options ...including none.

At first I was trying to get an asymmetrical filter to work out.  I was trying values between 17% and 5% pre to post ratio.  And while an asymmetrical filter can sound good,  the best "new AI" settings I've found is for a full minimum phase filter with straight TPDF dither.  The settings for this are 0% and then selecting TPDF from the dropdown menu for dither.

I wish the divisor for the filter bandwidth were active,  I'd try opening it further if I could.

And yes for dither straight TPDF is sounding better than any of the noise shaped variants.

[PeterSt]

I wish the divisor for the filter bandwidth were active

I am sorry, but what do you mean - "divisor" ?

[charliemb]

I wish the divisor for the filter bandwidth were active

I am sorry, but what do you mean - "divisor" ?

Hi Peter

Sorry for the delay.  Busy week.

There's a bandwidth or slope setting,  I don't have it in front of me right now.  As I recall it shows Fs/2.  It appears to be clickable with a drop down, but it isn't.  I guess this is grayed out.   I was hoping I could change the value to something like Fs/1.8.   

What I'm really trying to do is open the slope or control the slope, either the starting point, the ending point, the rate, or all three.  One interesting one would be starting from 16.3K and ending at say 22.05.   Another would be starting at 20K and ending at 24K or 30K (under the assumption that the original AA filter at the mastering stage did its job, as an option anyway).   

Why?   Because to my ear 48KHz Fs recordings sound so much better than 44KHz Fs recordings.   48 can sound almost as good as 96 or 88, just shy, or sometimes indestinguishable.

And okay, other ratios may be better than these.

I acknowledge that "opening" the filters does not always work.  For example, DSD sounds best with the lowest setting in my DAC, denoted by "<50KHz."   Meanwhile the chip default curves are 50, 60, and 70 KHz and none of these are acceptable.   One reason is that the quantization noise in DSD is full scale.   Whereas in PCM the quantization noise is over the last few least significant bits.   It seems that PCM has the advantage for this type of added noise.

[charliemb]

This happening now fairly consistently and therefore often enough to report.  Regarding the newAI vs. the 176 custom, I don't know how this is possible but it appears that the 176 custom filter at 352 can have more buzzing / zooming than the newAI filter using 0 (min phase) and TPDF as its variables.   This is not supposed to be the case.   Very strange.

As always,  ArcP none, or whatever is in the recording.

[BTW,  if it ever mattered, only now is my Sig remotely close to what it has been for months and most of this thread,  except that wasapi is new within the last few weeks.  The one thing I'm quite certain is inaccurate is the Q5 setting; it might be 5]

[PeterSt]

Hey Charlie,

Just to let you know : always reading.

I must say though - it gets quite messy for me. I mean, not at all that you don't make sense, but it becomes mighty difficult for me to make sense. There's just too many things different from what I do or have set;

The first example would be the OS which is W7 in your case and to which I most definitely can not listen any more (just one pile of distortion).
And if you *then* start about "more buzzing" with the custom filter opposed to a most-ringing setting in the AI filter ... well you wondered yourself of course, but how to help out now ...

Something else - also not making it more easy - I myself am back to normal Arc Prediction. Maybe not forever, but at least since 3-4 weeks by now. The "whether needed" (Custom) seems to depend on external factors I am not conclusive about yet. But an example : change interlink which does not work out until you move to the other filter.
What's possibly going on here is that "illegalities" in the filter (AI being the most legal of them) *or* show, *or* are filtered by the interlink. Or the other way around : when filtered too much it may require normal AP to push back some of the (square) detail.
Just far away thinking; nothing formal.


Peter

[charliemb]

A simple hypothesis would be interaction, ..with the ringing already present in the recording.  

And so the most natural sounding interaction is the one that goes unnoticed.

This would also explain why it is so difficult to pick a favorite filter amongst, for example, Miska's 15 or so filters in HQPlayer.  All sound different, and all eventually lead to a situation where a change to a different filter improves.