New Filter request(s)

[charliemb]

Non preprocess would be great because of load times.   Quick like AP or no filter.

For us in the 4x world (192 DACs), perhaps it might be a good idea to give us a choice to preprocess AP, AI, and this new filter.   We won't have the huge files of double octo.  I say this because  AI is coming in ahead of AP right now with DAC almost wide open (slow filter, unlike the setting in my sig).   The order, best first is:

no filter @4x,
AI @4x,
AI @ 2x, and finally
AP @4x.

But AP should win overall or at least any AI.   AI is a preprocess.  So I can't help think that if AP were also a preprocess (optionally chosen in settings) that it would win out.   

[PeterSt]

A few updates ...

On of them is that only after a longer while I found that this all even can't work with tracks larger than around 8 minutes becaue that implies more than 2GB of file while WAV officially can be 2GB only. That all goes quite unnoticed because nothing complains really but some sort of random noise exhibits once the 2GB limit is passed during playback.

WAV can be larger with some tricks, but in the end this is not the way to go; I got it all working quite well and without lag when playback is started and by now a 12 core (or more) processor really is convienent; loading of an album of 12 tracks or less takes maybe 1 minute but in the end as long as the longest track. However, would the album consist of 1 track of 60 minutes, it theoretically takes 12 minutes.
Playback starts right away anyway, but it degrades the SQ performance hence is against all what XXHighEnd is for.

So, not the way to go.

[PeterSt]

And so I finally sat back, read into a few basics and started to build my own filters right from the ground up. Or at least one type (many exist, each with numerous options).

Small side note : I never approached it this way because it was my thinking that only Arc Prediction would be lean enough to perform it as a real time process (and then with parallel processing harmless to SQ). But since "back then" quite some things happened and one of them was learning how a processor speed of 430MHz would still imply a virtual 0% CPU usage, also because of improvements over time. And of course now encouraged by the supidly super large files otherwise (pre-process) I just had to give this a go.

Well, I should have done it earlier. Or at least I just as well could have;

Of course all starts out with selecting a filter type which doesn't take ages for its necessary processing but which is to combine with the desired result. So, that already implies quite some "reading into", also (by now) knowing that it takes several days of programming before something is working to begin with, to only see then the result. Thus that is no way to travel the paths and the theories must be well known in advance. Takes additional time, but is gained back right away because the filter should work out as intended if only no programming mistakes have been made. And I can tell you, this is tough already. Example :

I ever back started out my programming life with creating a chess program. This was cool, and fairly quickly I could not win of it myself. Small problem : I could see that it would let trade her queen for some minor piece and I could never see why. I lost from the program all right, but there had to be a major flaw in it anyway.
Now, such a thing can not be debugged; what such a program does is "trying" all the moves as a first step and look what the opponent (me) would be able to do for each of such a move. But for each move of the opponent again it analyzed what the program will be able to do. This (as I recall) 3 steps deep from each side, or otherwise there wasn't enough (processing) time. Uhm, back in 1976 or so.
Anyway - and I forgot the exact numbers - when you're in turn you will be able to make, say, 200 moves. The opponent the same for each of your possibilities. This 6 times gives 200x200x200x200x200x200 = 64,000,000,000,000 settings which are analyzed. In at least one of all these settings a queen is offered for the better, but since this is about the stacked result of all there is no way of debugging this. It is too much to begin with.

The chess example would still lead to discrete analysable results per setting. Not so with audio (data). You can't even judge the input, let alone what the filtered result of the output has to be. It just has to be 100% working OK without a single bug.
And then to think that I saw example code (lectures actually) which in practice lead to something like SQRT(-1). Well, try that. It can, it actually should for this kind of filtering which is about real numbers and imaginary parts, but if the lecture code already fails ...
And you know what ? I found all kinds of real programs which worked with that lectured code ...

Right. Didn't I tell that existing filtering means measure worse than Arc Prediction for the lower frequencies ? I did. But now I made a first one my own, and that measures BETTER.
Aha ...
Well, marginally, just as the others measure marginally worse, but at least it gives me the idea that I possibly did better (somehow).

Not unimportant : When I run this on top of Arc Prediction it is not even noticable (for the time it takes). Or not yet, because I now firstly use a most simple form (though that's not for complexity or speed reasons but for not being a steep filter - maybe the whole subject of this topic).

Now my (or "our") major advantage is that I can look at Arc Prediction and know what's to be improved on it and especially what's to remain. The latter of course is about the transients, plus AP shows how phase behavior should look like. So all I need to do is look at the latter two while tweaking the fomer to improve. Actually quite easy ...

And so I now listen for the past few days to a first theoretically set filter which looks good and it sounds better.
Aha.

"Sounds better" could be subjective because sound does change and what the brain needs to do is whether changes for the better do not get overwhelmed with things for the worse. So, to me and for now it seems true that after listening to this filter for a couple of ours and then switch back to AP, it sounds more accurate again but then the other way around shows more "music" right away. So let's say a more analitic sound changed into a better coherent music sound.
But one super major thing : I can't recognize a flavor of the filter up till now, let alone that I get crazy of the zooming etc. like commercial filters do (e.g. iZotope) no matter the settings.

Notice : In a previous post I talked about the hardly visible pre ringing and 4 samples of post ringing. That (preprocessed) filter worked quite well too, but that was still too obvious for the smearing; I envisioned myself making it up but with *that* filter I already thought "oh well, if some people like it for the better they go ahead with it". Not so with this real time filter; I think I really like this one for the better myself. But then this one is phase linear and meanwhile it doesn't ring either. Well it does, but to the same extend that other filter pre-rings (hence hardly visible).

So far for now.
Peter

[charliemb]

... So let's say a more analitic sound changed into a better coherent music sound. ...

Thanks  Peter.   That's exactly what I think I need in my setup  --that choice.

But one super major thing : I can't recognize a flavor of the filter up till now, let alone that I get crazy of the zooming etc. like commercial filters do (e.g. iZotope) no matter the settings.

I don't know what is meant by "zooming."

[PeterSt]

I don't know what is meant by "zooming."

LOL. And then to think I use that word weekly for years before someone tells me it doesn't exist.

Of course it is totally obvious to me now that this is in-existent. We Dutch ... We say zoemen (pronounce oe as oo).

Buzzzzzz

But buzz with oscillation like a fly sounds when flying.

Then of course people from the northern part of Holland should be careful because what normal Dutch call a fly, they call a mosquito. And what "we" call a mosquito they call a fly.
I don't know how *that* happened.

[Stanray]

I don't know what is meant by "zooming."

But buzz with oscillation like a fly sounds when flying.

How does a fly sound when NOT flying?

 

Stanley

[charliemb]

Hi Peter

Gee...  How much longer are we going to have to wait for these new filters?
I've been checking daily...  it's driving me crazy a little.

Charlie

[PeterSt]

Haha, that's your own fault, because you don't have a NOS1.

Well, sort of and not really of course, and the problem is merely this :

Without first making *that* we have no provision of seeing what the result of the filter will be. So this is in another leage (if emulating it with plots will show a fair picture at all and I do NOT think so), and this is what has held me back somewhat. Ok, apart from the time it takes while clearly some other things were (and are) going on in parallel. But think of it : you won't be able to "see" what the filter causes, nor will I myself. And this latter I can at least do for the NOS1(a).

Anyway, see beta-beta download below;
You will see quite some improvements for the AI filter which even to my ears is listenable but which I gave up upon because of it's (still) akward-ish usage, plus no real solution in the end. Nice for the Frequency Domain, but that is really all and it will never serve the Time Domain.

So (and already talked about I think), in there there's also a "Custom" filter, and *that*'s what the sh*t is all about. Next small problem (no, same problem) : although that works perfectly all right for the NOS1(a) it can not work the very best for 176.4/192 only (44.1/48 upsampled). This is just too low to avoid images in the audio band (and the Anti Image filter does not).
Still I made one for you (or anyone else with 192 max capabilities) so you can try. And it should at least work way better than normal Arc Prediction. It still does not ring a bit, and "utilizes" the frequencies we won't perceive anyway, assumed we are old enough to begin with. Just as you asked for (the new AI filter in its default settings does that too but keeps on ringing, but less now).

For NOS1(a) users ... they too can try the AI filter of course, but they for sure should try the "Preferred" 705600 Custom filter (you will understand when using it). I am using that myself for maybe 6 weeks now, and I don't think I will go back. Improving it yes, but I never spent a second on it after this first attempt.

People should notice that the Custom filter's name is exactly that : ready for you to set in all directions. But back to the beginning of this post - you won't be able to see what you do, and it's really "no task" to start dialling without analyser (or well done emulation of which I think it is not possible).
So what I'm saying is that he "customizing" will be for myself for now, and Pre-selections will emerge from it. Just like how it is now (with actually one setting in there at this moment).

The settings are envoked by means of the [ F ] button you see below (see mouse arrow) and the Custom Filter is engaged by means of the "Custom" button above "ArcPredict" (top left section). Don't forget to click the Help link in the settngs screen.

The download at the bottom contains 4 files. Unzip them over your current 1.186 or higher folder and nothing else (!). Before doing so, kill XXEngine3, XXHighEnd and AutoHotkey from memory.

Important : This version will not allow you to go back to the ever existing version of AI filtering; for that you'd need to revert to your current XXHighEnd.exe and XXEngine3.exe. So set those aside when you think you want to compare.

The 1.186-g you will see is just the latest version but no Release Notes. All I can say is that it may be quite useful for (tablet) remote operation (since I started to use that exclusively myself again, haha).

For all who try this for the filter, let me know your findings !
Peter

EDIT : At this moment it appears so that the Custom Filter does NOT work for "DAC Needs" = 24 (bits).
This is not sorted out yet, but take it this implies static (loud noise). See posts below.

Edit June 24 : Removed the download because of a few issues. A new download (1.186-h) will emerge in this topic soon.

[dsm]

Thanks Peter,

The custom filter does seem to be a noticeable improvement on my non NOS 1 system. But a word of caution,  I was using AP upsampled to 192 and left that setting when I switched to the custom filter and received  deafening static. 
All works well without upsampling and this may just be my system but others might want to pay attention to the upsampling setting.

Regards,

David

[PeterSt]

David, did you ever enter that [ F ] screen first ?

Are you 100% sure you removed XXEngine3.exe from memory ?

What did you select in that [ F ] screen ?

All IOW, something is not right and nothing of the sort should have happened.
Similarly you can't even use "no upsampling" at all ... (it will give you an error message that Filter 5 could not start).

So I'm a bit clueless.
But sorry that this happened to you. Now let's (seriously) work out how.

Peter

[PeterSt]

But for a next one trying ...

It could be crucial to first go into that [ F ] screen before really playing; that should fill all the data, especially when nothing has been selected yet from the "pre-selections". Now, it is so that I obtained a provision for that, but I never tested that really.
So David, if you can confirm that you just pressed the Custom button and never went into the [ F ] screen before that, this probably would explain it. But otherwise ...

??

Peter

[dsm]

Hi Peter,

Yes,  first went to the F  screen and chanced the setting to the 176/192 setting before starting the engine.   I am using an Audiophileo 2 with purepower at 24 bits fed into a marantz dac.
The static that is created when upsampled has a "pulse"  to it or a regular rhythm.  The original  music is not heard at all. 
I do get the warning about the filter sampling mismatch if I start from 48000 and go to 96000 but it I bump it up to 192000 then I have static with no warning.

If there  is something else you would like me to try,  let me know.

Best,

David

[PeterSt]

Hmm ... Are you saying that you have "DAC Needs" set to 24 (bits) ?

Thanks,
Peter

PS:

All works well without upsampling

This "message" I actually don't understand then. Okay, maybe you are *not* saying "works for the better !" but that is what I got from it, at least at first.

[dsm]

Yes,   my dac needs 24 bit in order to play.  Now,  I may be a bit ignorant about some of this but does the custom filter make any difference without upsampling?  With the quick listen this morning,  I could have sworn there was a little bit of an improvement in the sound stage.  Am I wrong? 
I've tried a couple frequency's and have trouble with upsampling any with the with the custom filter.  Some will give me an error and will play normally once I close the window about filter 5 not starting,  but others will just be the static.

Regard, 

David

[PeterSt]

David,

When the output sampling rate is lower than that of the filter chosen, the error appears (should appear) and no Custom Filter will be engaged. I think that Arc Prediction will be engaged instead but I'm not even sure about that now (actually it should not play at all, but beta-beta ...).

So it should be about the 24 bits, but to be honest I never even thought about that while I also can't even see how it would be related at this moment. But let's take it it is.

Sorry David,
Peter

PS: I put a message about this in the "OP".