'FeralA' decoder -- free-to-use

[Jud]

1 hour ago, John Dyson said:

When doing the FA decoding, in previous versions, the vocal was 'buried' deeper than it should be. 

 

Something I've noticed and had mentioned to you. Glad to hear progress is being made in that area.

[John Dyson]

6 hours ago, Jud said:

 

Something I've noticed and had mentioned to you. Glad to hear progress is being made in that area.

I really, really do listen to peoples  comments.  The 'buried vocals' have worried  me also, but I couldn't see a solution.  I couldn't dignose the problem until the last day or so.   We could all hear a problem, but the problem needs to be deeply understood before prescribing a solution. 

It has been *infinitely* frustrating to be unable to hear the structure of the defect so that it could be diagnosed in enough detail to solve the problem!!!

 

A very focused discipline needs to be maintained so that I can finish the project.   When listening in great enough detail to diagnose the signal defects, my hearing easily becomes strained.  Something goes wrong, and it appears that I have AT MOST 5 minutes of listening time before the ability to hear details (and tells) will disappear.   This hearing defect does NOT match my work-ethic and innate nature of working super-intensively.

 

Adopting this more relaxed way of working has enabled better signal diagnosis, therefore better progress.   This new process requires discipline to listen only in short intervals,.  Being more 'careful' has enabled much better productivity and better ability to hear all of those 'tells' and 'details' that I used to be able to detect so easily.   I sometimes fall into a trap of enjoying  a recording, and listen for longer than a few seconds.   Taking the luxury of listening to a whole song will disable progress for a big part of a day.   If listening for enjoyment longer than a short interval, I must back-off for an hour or more, because I will have become unable to diagnose the signal defects.

 

Not just the buried vocals, there have been several other major problems that became solvable by avoiding a 'hearing burn-out'.   I don't know what is going on about my hearing, but  It doesn't seem like it is a 'mental processing' fatigue, but instead my hearing isn't able to withstand a certain kind of strain.

 

I do THANK YOU and others who have made useful comments,   The only change has been in the work-process,   The feedback has ALWAYS been appreciated.

Imagine the frustration!!!

 

 

 

 

[Skip Pack]

One thing that popped into my mind was a suggestion for a new category of  test music. In the snippets, at least, I don't hear and example of solo accoustic piano with the long-tail reverberations that a piano is capable of producing with it's tonal range and pedals. 

You may well be using selections of this music but not including them in the snippets we hear, but It seams to me that you this music may illuminate the effects and balances you are addressing at this point.

 

If you think such example might help, those of us following your efforts could suggest specific performances. I would expect examples from the '70's or so that were popular enough to get mass-market CD releases in the 80's or 90's.

[John Dyson]

On 5/2/2021 at 1:28 PM, Skip Pack said:

One thing that popped into my mind was a suggestion for a new category of  test music. In the snippets, at least, I don't hear and example of solo accoustic piano with the long-tail reverberations that a piano is capable of producing with it's tonal range and pedals. 

You may well be using selections of this music but not including them in the snippets we hear, but It seams to me that you this music may illuminate the effects and balances you are addressing at this point.

 

If you think such example might help, those of us following your efforts could suggest specific performances. I would expect examples from the '70's or so that were popular enough to get mass-market CD releases in the 80's or 90's.

Okay -- There IS a new V2.3.1A uploaded today (yes, the version upgrade shows major improvements -- I hope.)   I haven't done the overnight check with truly fresh hearing yet -- there might still be a little bit too much of HF, but I cannot tell.   The HF can easily be adjusted, but any HF error at this point is a matter of dynamics, shouldn't be the technical gain level. This is NOT an offiical announcment, just a heads-up that the files are ready -- IF the results are as good as I think that they might be.

 

Also, I'll add some snippets with more'piano' stuff -- I'll be sending you a private message tomorrow about some other matters about the decoder.

 

The new piano stuff isn't up yet, and the snippets are currently the same as before.

Tomorrow, I'll do the official announcement once I do the review after very rested hearing and will select/add the snippets a little later in the day.

 

This new release snould a MAJOR MAJOR MAJOR improvement, not just about 'signal levels;'.   The 'secret' the LF has been found, and the stereo image/vocals have been brought more forward.  The designers of FA did a really evil phase scrambling and phase shifting scheme...   Really, really subtle stuff is needed to clean-up the audio.   I would have NEVER figured out the crazy phase thing when assuming that my hearing is okay and using my old development process.   My hearing is just not very good at all, and has very minimal accuracy for only a few minutes.   It really requires discipline to avoid listening more than a few minutes, when my style of working is so very intense!!!

 

*  I really, truly depend on 'tells' and subtle kinds of distortion for knowing the changes needed.   The phase issues became clear after the hearing matter was understood, and know that there are very limited circumstances when my hearing is of 'audiophile quality.

 

Let me know what you think about the qualities of the test stuff  -- publically or privately.   I *know* that it is very very close now, but need some people with good hearing to give the final bits of feedback.   There IS a V2.3.1A decoder (fixed) uploaded, and shouldn't have the math lib problem anymore also.

 

Thanks for the piano suggestion!!!

John

 

[PeterSt]

The last I listened to was 2.2.9S-0; my comment below thus only reflects that or earlier versions:

 

On 5/2/2021 at 1:13 PM, John Dyson said:

I don't know what is going on about my hearing, but  It doesn't seem like it is a 'mental processing' fatigue, but instead my hearing isn't able to withstand a certain kind of strain.

 

John, it is so wrong that no brain is able to cope. This is except for 2 or 3 versions and along the way I told you about those. But you seem to go your own way and instead of using those versions as a base and work relatively to those, you implement new ideas and it drifts off (wildly) again. N.b.: You did not receive comments about 2.2.9S-0 and a few more, because you explicitly asked not to.

 

Yes, I will bet you that you are quite explicitly destroying your hearing with this; Your brain inherently will know how music is to be (it will especially know about human voices) and listening to e.g. 2.2.9S-0 is plainly impossible for me. It is literally hurting. But contrary to you, I will stop listening after 10 seconds of two tracks (of either Supertramp or the general Demos) and quickly escape. Read: from natural reasons I will not allow further damage. I = my brain. I quickly recover - you can't because your urge is in listening to the results.

Also the fact that you are dependent on more than one person for aiding your hearing, should be killing as such already. It is like your three best advisors and friends giving you three totally different options to choose from while one option is allowed only; you will be literally in circles between the three and the necessity to choose is straining already (and from an other angle than the hearing issue).

 

---------------

I am actually responding to your quote only, which is obviously what you came up with yourself and which won't be without reason; Unconsciously you will feel that something like that is going on but you refuse to accept that (another straining position).

---------------

 

Later today I will listen freshly to 2.3.1.A-0 and report by PM.

Peter

 

[John Dyson]

4 hours ago, PeterSt said:

The last I listened to was 2.2.9S-0; my comment below thus only reflects that or earlier versions:

 

 

John, it is so wrong that no brain is able to cope. This is except for 2 or 3 versions and along the way I told you about those. But you seem to go your own way and instead of using those versions as a base and work relatively to those, you implement new ideas and it drifts off (wildly) again. N.b.: You did not receive comments about 2.2.9S-0 and a few more, because you explicitly asked not to.

 

Yes, I will bet you that you are quite explicitly destroying your hearing with this; Your brain inherently will know how music is to be (it will especially know about human voices) and listening to e.g. 2.2.9S-0 is plainly impossible for me. It is literally hurting. But contrary to you, I will stop listening after 10 seconds of two tracks (of either Supertramp or the general Demos) and quickly escape. Read: from natural reasons I will not allow further damage. I = my brain. I quickly recover - you can't because your urge is in listening to the results.

Also the fact that you are dependent on more than one person for aiding your hearing, should be killing as such already. It is like your three best advisors and friends giving you three totally different options to choose from while one option is allowed only; you will be literally in circles between the three and the necessity to choose is straining already (and from an other angle than the hearing issue).

 

---------------

I am actually responding to your quote only, which is obviously what you came up with yourself and which won't be without reason; Unconsciously you will feel that something like that is going on but you refuse to accept that (another straining position).

---------------

 

Later today I will listen freshly to 2.3.1.A-0 and report by PM.

Peter

 

You might want to hold off for the 'E' version.   There were some problems with 'pianos', and too much high end.   After a full nights rest (about 4Hrs), my hearing is temporarily

restored, and I had realized that there was a few dB too much on the highs (yes, it doesn't take much error for the problems to become apparent.)   I did mention that the version online wasn't really a normal release, just informational.  The 'E' version will probably be an announced release.  Also, in the 'A' version, the 'dynamic' lows were'n't tracking correctly.  The LF EQ needs to change  beyond the normal DolbyA actions, but wasn't wasn't tracking correctly in the 'A' version .  (only one of the excess dimensions of control.)   This tracking/curve shape error made the pianos sometimes muddy.  There are three common  interactions to correct the  LF at the lowest and highest gains.   Some of the EQ changes are only indirectly related to both the LF/MF gain straight through the system.

 

About tracking or following up on previous versions,that might be true but the decoder is like a 4D tetris game -- things don't work in a simple linear way as in normal day-to-day

interactions with technology.  Wierd interacting  and strange side effects make 'following' or 'building' on a previous version being only partially useful.  There IS some use by looking at history, which I very much do, but the variables are almost NOTHING like tweaking frequency response or EQ,  not even as a serial set of EQ like a parametic or other common EQ.   This complexity is probably one reason why many people have been frustrated -- believe me, it is was as simple as it might seem, the EQ would have been correct

very quickly.   A few of the 'dimensions' is associated with the architecture itself, which now appears to be correct.  Some aspects of the architecture cannot be easily determined

without hearing  the subtle distortions from gain control dynamics.  I have REALLY learned a lot and appreciate that aspect of my hearing that is still somewhat reliable.

 

* The notion of 'tetris' is actually more true than one might initially believe.   The changes are 'blocks' and not 'tweaks.'

 

I understand the notion that there isn't 'building' on existing EQ, but  there are multiple lines of EQ which mutually interact, along with 'borked' hearing.   I would probably have the same thoughts if not actually doing the design -- and visualizing the changes.   It is at least 2D + a few additional 1/2 dimensions of EQ.   Multi-dimensional visualization is tricky at three dimensions, and bording on crazy beyond that.  More limited multi-dimensional thinking is also needed on the anti-MD. Picturing all of this stuff is mentally exhausting, and already am very fatigued and burning out because of the attempt of working very intensively to divine information that is made unreliable by variability of hearing..  By the time a change is made of affect an audible change in response, there are often 2 (really two, perhaps three) variables that also have to change...   Those changes are not always in the common-sense direction.  As a simple example, in one example location, to soften the dynamics in the super-highs, some of the EQ has to INCREASE the strength of the super-highs.  Of course, it DOES make sense once understanding the shape of the curves, but chasing all of the variables is not trivial.

 

About the listening fatigue -- it is really decieving.  I always had some aspect of fatigue, but now this variability is really extreme.  This hearing problem is the main reason why there were complaints of 'house EQ', but instead I was probably perceving something close to correct -- just going nuts with incorrect information, often flopping around along with trying to remember the complex context about previous settings.

 

Thanks for your continued feedback -- this stuff is as far from trivial as I have ever dealt with by myself. (As design, of course people are trying to help with observation.)

 

 

 

[PeterSt]

15 minutes ago, John Dyson said:

As a simple example, in one example location, to soften the dynamics in the super-highs, some of the EQ has to INCREASE the strength of the super-highs.

 

FYI (but probably known already), higher resolution (is theoretically more highs (EQ)) softens the dynamics. This is always the case.

In the end this is dangerous because it requires speed of the system. If the system can't cope (on e.g. slew rate) then you're only implying distortion. So ... you may also wonder whether your system is on par to begin with. Read: the better you make the resolution the more it requires speed, the more distortion will be your share when the latter is insufficiently present.

And the other way around : take out all the transient stuff and the system won't add distortion. However, now it will sound "dead".

 

There are so many reasons why this can't be done by listening. It can only be done that one and only good way.

 

N.b.: Most of the stuff I listen to, is the highest transient possible (I do this to make the (commercial) system sound the best it can, so the highest transient material is the best material for "testing"). While this material sounds the most holographic and esoteric or theatric or whatever special, sounds like total rubbish on other averagely good systems; they just don't have the speed to follow, or mush the resolution by (inaudible) noise otherwise.

 

I know how difficult it is to be at the really highest level of detail which requires the lowest level of noise to begin with; do one small thing out of order (like a playback buffer size) and it sounds nasty. Read: the balance between good and really bad is ever so small/fragile.

[PeterSt]

31 minutes ago, John Dyson said:

About tracking or following up on previous versions,that might be true but the decoder is like a 4D tetris game -- things don't work in a simple linear way as in normal day-to-day

interactions with technology.

 

Not important, but possibly funny:

 

Those who create sound with synthesizers (I do), will know how ever so fragile modulation changes will change the sound drastically. And, where you may not perceive a modulation at a higher frequency, it may be profound at a lower. And the other way around.

 

So what you are (un)doing is modulations of several sorts and they are stacked.

 

I have already been thinking what would happen (for beneficial result) if you could provide the "modulations" in a stacked fashion in a .WAV file (MP3 would be better, but might technically fail), so I could apply that file as a convolving file on to played notes with (even gliding) pitch and volume. Or even more practical:

Have two wave files, one with original CD music and the other with your convolving elements, and morph the two together until something overshoots. From there, there's this idea :

 

Have each of those elements in a separate convolving file, and dial it in for level with originals. Happy ? then fix that with the original, and play that back as a next original, to be morphed with a next element. And so on, as many times as you like. And at playback also in the pitch you like to hear.

 

I would be my idea to observe each of the elements (them being as raw/native as possible) for faults. I think this can be done (per element I mean) because it is my way of working at "system development" for audio too. Thus, I am able to hear through the faults elsewhere, to be covered for later. One element at a time, is the devise. Btw, you can see that happening in my "critiqueing"  (PM) so far; lift out the good elements, tell you to not touch those, and at the very end of the report, tell what to relatively change.

It is only that you don't do that, thus far. Hahaha.

 

John, IMHO there have really been some pearl versions already, those being better at aspects than the originals (from CD). But they are rare and same to emerge accidentally (no matter how much you want them to emerge purposely - but you should not care about that).

 

[John Dyson]

Some kind of automatic math mechanism might be helpful as you suggest, but the reference for the feedback is a problem.   The metrics would also be tricky.  I have been researching the math/dynamic measurement/estimation methods for detecting the very slight 'signal' when trying to sense FA or not on a recording.   The saving grace for the detection is the multiple layers, and there is a 'DolbyA' glitch on each layer.   The 'glitch' tends to be stronger on layers 2,3,4 and sometimes 6 (out of 7.)   After studying the methods, it appears that some kind of Kalman filter will be needed, but the nonlinearities will make it necessary to have some kind of preprocessing before using the Kalman stuff.   (Usually, such filters are used for integrating different signal inputs along with estimating the associated dynamics characteristics.)  A straight forward filter + threshold is probably not sensitive enough.

 

Remember when I mentioned the multi-dimensional aspect?   The dynamic LF response isn't working quite correctly in the V2.3.1E version.  The lows are pretty good in most cases, but

'ONJ's Soul Kiss album bass isn't quite right.   I need to understand the frequency of the bass, and then modify the dynamic LF stuff.  It is a tedious balancing act because right now, I am super proud of the relatively good sound of pianos on real classical recordings.   The problem with pianos is that the LF is tricky, the line between 'muddy' and 'correct' is very very thin.

 

Gotta be very careful when modifying the bass dynamics.   This 'bass' response is not directly related to the 'bass' response, but there is a nonlinear interaction.   The previous very 'muddy' bass (when there was bass) came from the total lack of the dynamic bass capability.   That was yet another 'block' diagram box that had to be added a few weeks ago.

 

Even though the real bass is approximately correct, or maybe even totally correct, the dynamic bass is still a little erroneous.   The 'dynamic' bass input is not an independent entity either, because it comes from the bass gain on the channel, but a more frequemcy selective reactive version.   The setting also interacts with the bass EQ.   There are similar interactions in the HF, but the HF is very very good -- perhaps needs some minor changes to level, but the sound is very appropriately 'sweet'.

 

So, there'll have to be an 'F' release.   As long as there is still progress being made, I do NOT want other people (users, testers, contributors) to spend time dealing with the errors.   I see/hear a solvable error, so going to move forward on an 'F' release.   There are SOME 'F' release examples online, but don't waste your time unless REALLY REALLY REALLY interested.

 

Why didn't I catch this before?   Of all of the recordings that I have listened to -- perhaps snippets on 100 or so, the 'Soul Kiss' is the only very significantly troubled bass.

 

 

[John Dyson]

Have a project for everyone interested.  The project DESPERATELY needs help/input with the best choice...

(Because I am asking for help, and I really need it -- I will be much more responsive than the last few weeks.  The only reason why I was 'hiding away' is a matter of focus, avoidance of distraction.   I had no intent to show disinterest in what users/contributors are suggesting.  In fact, I am VERY interested in suggestions!!!!)

 

I just uploaded the demos from the most clean decoder yet.   However, here is the gotcha -- there are three (3) candidate decoders, each with a profoundly different low-end.   Two have the same kind of low end, but one of them has 1.5dB less.   The third implements the low end with the same general  EQ structure, but with different constituent  1st order EQ types.   The previous versions had a 'perverse' LF/lower MF.   I had a terrible time creating a good sounding version with the older methodology.   The NEW methodology for these test choices below are actually my original method, but my bad and unreliable hearing incorrectly told me that the LF version used on the previous test release sounded 'good'.   I admit that the method used on V2.3.1 prior to the V2.3.1H variants could be made to sound reasonable, the settings were fragile and the associated mix with the midrange and upper frequency bands sounded uglier than ideal when critically listening. (This ugliness is the reason when I deferred the previous release attempt.)

 

Here is the test info, and the location of the tests.   If you look at filenames residing in the Dropbox location, you'll see V2.3.1HA, V2.3.1HB and V2.3.1HC.   I ask that a simple comparison be made on several snippets of your choosing -- which version is better?   When I get the information from you, the decoder will be reproduced and released with the techniques used in the selected version.   Here is the location, and below are more details about what is going on in the decoder (optional reading):

 

https://www.dropbox.com/sh/i6jccfopoi93s05/AAAZYvdR5co3-d1OM7v0BxWja?dl=0

 

THANKS FOR YOUR HELP.   Feedback is welcome both publically and privately.  All I really care is that you provide YOUR opinion.   I truly do not know which version of the LF range is actually technically correct.   If I needed to choose, it would possibily be V2.3.1HA -- but could just as easily be any of the other two.

(The binaries for each version of the decoder are also provided in a subdir of the Dropbox location above.)  One of the versions from the selection IS technically correct or very very close.

 

------------------------------------------------------------------------------

Optional verbosity:

The odd-man out on the 'different' 2nd order EQ type uses 2 each of 3dB 1st order EQ, the other two uses 1 each of 6dB 1st order EQ.   Note that even though the EQ value is the same between using the 2/3dB EQ and 1/6dB EQ, the shape of the response curve ends up being very different.   Using the 3dB steps will provide a bit less lower midrange.  Using the 6dB step will create a more full sound...   One needs to be careful when choosing between the EQ types because 3dB EQ types will give a more 'detailed' kind of sound, but such 'detail' is a double edged sword.   The 6dB EQ type can give a sense of less distortion, but is very easily capable of producing muddy sound.

 

The big difference between this version and previous is that instead of additional 2nd order EQ (about 9-11dB is needed), an additional 4.5dB of 2nd order is added also.   This 4.5dB offers a quicker increase below 75Hz, and better matches the 80Hz band response shape of a DolbyA unit.

 

So, there are three versions of each recording.  I am not asking that ANYONE listen to all recordings and all three versions.   All I ask is to listen to several selections that you might know well.   The names of each version have a version ID like 'V2.3.1HA', 'V2.3.1HB' or 'V2.3.1HC'.

 

I had worried about telling everyone which version has what kind of EQ, but I don't think that it produces all that much prejudice, so here it is:

 

V2.3.1HA:   uses 6dB of 1st order, 4.5dB of 2nd order 75Hz, Q=1.414

V2.3.1HB:   uses 6dB of 1st order, 3dB of 2nd order 75Hz, Q=1.414

V2.3.1HC:  uses  3dB+3dB of 1st order, 4.5dB of 2nd order 75Hz, Q=1.414

 

As you can imagine, these are not the complete set of choices, but any choice of the three, plus additional comments about deficiencies will help me modify one of the above to be the final version.   My own choices are V2.3.1HA and V2.3.1HC.   'HC' has a more thin lower midrange, while HA has a fuller midrange.

 

PLEASE give me an idea of what you think.   Here is the location of the choices:

https://www.dropbox.com/sh/i6jccfopoi93s05/AAAZYvdR5co3-d1OM7v0BxWja?dl=0

 

Again, thanks for your feedback.

 

[semente]

John, what you are asking is even more difficult than giving directions to a blind man driving a car, over the phone. 😞

So many variables, which keep changing from version to version.

[John Dyson]

3 hours ago, semente said:

John, what you are asking is even more difficult than giving directions to a blind man driving a car, over the phone. 😞

So many variables, which keep changing from version to version.

Hehehe -- that is exactly the troubles that I sometimes have---    I am mostly asking for some opinions on these need to make an all-or-nothing choice, but I just want an outside opinion on the *qualitative* differneces, not on *quantitative* that one or the other is absolutely the best.  Since the decoder is now close to correct -- choosing the best is becoming more and more difficult.   At the risk of bias, I believe that 'HA' is the most technically correct, but is NOT entirely correct.   The final version will have more low bass.   If I get a large percentage of people who prefer the sound of 'HC', then I need to revisit some things.   'HC'  can sometimes sound like it has 'more' bass, but that small deficiency in HA has been resolved in the current research version.  The primary difference between 'HA' and 'HC' is a sense of 'coherence'.   I want to know if people prefer the 'softer' since of HC or the more 'tight'  since of HA is preferred.

 

BTW, the current state of the decoder, on the HA version specifically, I have been able to increase the bass a little bit (legally, following the rules) -- the amount of bass is close to HC, but continuing the 'clarity' and 'coherence' of the original HA.

 

Here is what I am *really* asking.   This is NOT about the amount of the bass, but can people hear the difference in clarity between the three?   The goal is 'best sound' along with 'best bass'.   'Best' is not always 'the most'...  (I know tha the term 'clarity' and to a lesser extent 'smooth' have a sense of prejudice about the choice, but please don't interpret that I definitely believe that one or the other is better -- this question in my mind is the reason WHY I am asking it.)

 

 

[rando]

3 hours ago, semente said:

John, what you are asking is even more difficult than giving directions to a blind man driving a car, over the phone. 😞

So many variables, which keep changing from version to version.

 

I believe this is what a younger relative is currently facing with education becoming a virtual snooze followed by homework.  Using a medium known to be disruptive to human patterns their teachers attempt nuance and polish they should be attaining using multiple choice mixed media quizzes.  🥺🤯  🙂

[Confused]

3 hours ago, semente said:

John, what you are asking is even more difficult than giving directions to a blind man driving a car, over the phone. 😞

So many variables, which keep changing from version to version.

With a bit of effort and focus, it is surprising what can be achieved sometimes.

 

 

[John Dyson]

11 minutes ago, rando said:

 

I believe this is what a younger relative is currently facing with education becoming a virtual snooze followed by homework.  Using a medium known to be disruptive to human patterns their teachers attempt nuance and polish they should be attaining using multiple choice mixed media quizzes.  🥺🤯  🙂

The reason for the question:   My judgement is somtimes questionable, and want to make sure that I know which direction to go in.

Per some suggestions, I am looking for some good piano and orchestral mix examples.  (definitely solo piano included.)   I have zeroed in on a few albums, but need to do some selections/sippets.    My archive hs the complete version of one item with a good mix, the hard part is finding the correct subset.

 

[John Dyson]

Just re-evaluated the high quality modes -- --xp (super quality) is now practical for faster maches, and --fx or --fz are pretty quick on any 4 core AVX2 machine.  This speedup came from optimizations, not quality loss.  (In fact, there might be slightly better MF quality.)   I had realized that the speed was too slow for high quality, and in fact might still be a little slow when compared with a 'desirable' speed.   The new behavior also decreases the desirability for running in DP mode -- and DP mode will no longer be the default for AVX512 machines.

 

Lately, I have been using --xp mode, which REALLY gets rid of a lot of the gain control distortion that is endemic to the DolbyA type processing.   The --fz=opt or --fx=opt help certain situations, but most often, --xp is best.   The best choice list (in order of increasing quality) will likely be (none), --fx, --fz and --xp.   For certain slow machines, --fx=plus might be

useful.

 

It is really nice to see the 'dots' going by so quickly now!!!

 

[PeterSt]

I did not have the focus to seriously observe and put down worth while comments (busy making a serious dinner), but the fact that I listened throughout Crime without ever thinking** to hop over to something better, should be speaking ...

 

**): with hunches at times of an "oh wow ?" ...

 

 

PS: V2.3.1H-0

[John Dyson]

New release:  V3.0.2H.

 

Decoder:

https://www.dropbox.com/sh/5xtemxz5a4j6r38/AADlJJezI9EzZPNgvTNtcR8ra?dl=0

 

Snippets:

https://www.dropbox.com/sh/i6jccfopoi93s05/AAAZYvdR5co3-d1OM7v0BxWja?dl=0

 

The snippets are uploaded also.   I also added some original snippets for comparisons.   Some don't show much improvment, some show serious improvment.   Very compressed, loud bass is common on the RAW versions, but the decoded is more 'tight'.   Sometimes, the decoded versions will have a lower general level at the start -- that lower level results from the expansion increasing the dynamics.   The dynamics are mostly expanded downwards, even though there are some elements of upward expansion.  

 

All snippets have had their rate converted down from 88k/floating point.   When a 'RAW' version is presented, it might be in 44.1k form or 48k form.  In both cases, raw and decoded, the signal has been converted.   No matter what, the biggest change is downwards from the decoded output of 88.2k/FP down to 48k/24.  (The achives are sometimes at the full 88k/24bits, never FP.)   When decoding, a reasonable estimate would be adding about 2bits to the signal quality.   SO, even the garbage 14effective bits with MQA can be recovered almost to 16bits of audible quality if the MQA damaged recording is FA.   This mechanism is a direct analog to the use of NR systems (DolbyA, DolbySR, DBX, Telcom-C4, etc) in general.  They mitigate audible hiss, effectively similar to 'dither noise'.

 

 

 

 

 

 

 

 

[jabbr]

I just tested the "1984 Whitney Houston" album ... the 3.0.2H decode is very pleasing, the background hiss is reduced, the voices are slightly more foward/present, and there's a touch less ambient air, which is exactly what I think you are going for.

[John Dyson]

2 hours ago, jabbr said:

I just tested the "1984 Whitney Houston" album ... the 3.0.2H decode is very pleasing, the background hiss is reduced, the voices are slightly more foward/present, and there's a touch less ambient air, which is exactly what I think you are going for.

About the ambience matter -- that is definitely frustrating, because I have run into cases where I prefer more ambience than what normally happens after decoding.   Part of the goal is to try to be accurate, but that also includes trying to avoid creating further damage.   It is pretty obvious that some earlier versions of the decoder DID remove more ambience than should have been. A high priority is to avoid this 'excessive loss of ambience', but bugs still might lurk.

 

When things settle down, we might be able to talk about possible options for retaining more ambience.

 

Before waitiing to discuss more complete details, here are some ideas:  A good try might be to use --fa=7 or --fa=6 instead of the default --fa.   Also, when there SEEMS to be an excessive loss of ambience, it might be from the calibration offset (threshold) being too high.  So, the default --coff=-2 might be suboptimal for that recording.  Maybe a good idea to try --coff=-4 instead.  (most of the time, recordings might need --coff=-4, the other most common value.)   If --coff=-4 doesnt' work, then getting out the 'big guns' by trying --fa=7 or --fa=6 might help instead.   BTW:  the default decoding mode is --fa=+7, which has a different ordering than the suggested alternative --fa=7.   For example, the Supertramp recordings require --fa=7 and --coff=-4.  The --fa=7 (or --fa=6) and --coff=-4 combination seems to be fairly common.

 

Oh well, there just might be a loss of desirable ambience on a proper decode also -- just wanted to give some options that JUST MIGHT work better...  These changes discussed above ARE canonically correct -- but sometimes even a recording that doesn't require a variant decode might benefit.

 

 

[John Dyson]

Got some really useful feedback about this recent release.   To those who have presented some ideas and made helpful comments -- THANK YOU...

 

There are a few problems, and corrective progress is being made:

 

1)  Bass needs some work -- I have a solution for the bass issue. (Already existent, but commented out in source code.)  If I find that some recordings need the correction, and some do not, there will be command option added -- but only if needed.

2)  There was some latent modulation distortion,  causing vocals to have excessive presence (it really is a modulation of the MF vocal frequencies with higher end backing and higher order vocal frequencies.)   This modulation distortion comes from a pre-emph/de-emph mismatch with the original FA encoder design.

3)  The incorrect pre-emph/de-emph also caused a problem with the dynamics of brass and other similar instruments.   It also can have an impact on the presence of certain

kinds of vocals.   With the pre-emph/de-emph correction, the dynamics should be stronger and a little more natural.  (That is -- less compression left-over from the FA encoding.)

 

So far, the above are the current impairments, will might possibly collect more over the next few days.  Each of these impairments is technically very minor, but very important when demanding usefulness from the decoder.  All three problems is solved by no more than correcting the pre-emph/de-emph equalizers, and adding a couple of LF rolloff equalizers.  BTW -- the LF rolloff equalizers already existed in the source, just commented out.   This is totally trivial, and part of the process of 2nd guessing the original designers details.  NOTHING so-far has been found to be architecturally wrong in the V3.0.2X versions.

 

The bug with the pre/de-emphasis is that my original versions (V2.2.6H, V2.3X and V3.0X are all different) was that the current versions were too 'light'.   The amount of EQ needed is actually stronger than the already strong EQ that was being used.   Instead of the simple 3kHz +9dB & 9kHz -9dB, it acutally had to be 3kHz +9dB & 6kHz +3dB, & 9kHz -9dB & 12kHz -3dB -- this set of EQ is on the pre-emphasis side, on the de-emphasis side there needs to be the mirror image.    This addition of the +-3dB with 3kHz offset on each side makes a profound difference in quality.   Believe me, I tried all sorts of combinations that were architecturally consistent with the underlying DA hardware, and this is the only combination that cleans up the vocals.   (This is only one VERY SMALL set of variables that needs to be determined in the architecture, let alone the architecture itself -- this has been a long, painful journey.)

 

For the current released version, you *might* want to add some LF rolloff EQ at 25 and 37.5Hz.   Depending on your listening environment, that might greatly improve the utility of the decoder output until the new version appears.   As you know, my hearing is sometimes blind to LF -- and apparently it has been.   So, the next release will be corrected at LF also.

 

I won't promise a release date/time, but would be totally surprised if the new version isn't available on/before Wednesday night (USA Eastern time.)   Even though the answers are well known so-far, the new release process requires great care to avoid creating more fiascos.

 

 

[John Dyson]

Quick note about the pre/de-emphasis correction -- even though the results are somewhat improved on many materials, the vocals are much more clean on ABBA now.   The vocals are less diffused by modulation issues.   I didn't expect this kind of improvement, but it is real.

 

[John Dyson]

The V3.0.4C release is ready.  I decided to stop deferring the release announcement for all preparations on my archive site to be ready.

The decoder is ready at normal distribution locations, the snippets are also available.

Some work still needed on my archive site, and all of the stuff won't be ready there for about 1 day - not of general public interest, so no reason to hold back the release for everyone.

 

The snippets are available in the usual place:

https://www.dropbox.com/sh/i6jccfopoi93s05/AAAZYvdR5co3-d1OM7v0BxWja?dl=0

 

The V3.0.4C binary is available in a subdir -- direct pointer is below:

https://www.dropbox.com/sh/5xtemxz5a4j6r38/AADlJJezI9EzZPNgvTNtcR8ra?dl=0

 

I believe that all known feedback issues have been resolved.   The most important basic usability matter (LF response) is fixed.

The LF fix has brought out a problem with recordings that I had hoped didn't exist:  Some have either had a small amount of bass boost

or bass cut.   In order to explain this, I plan to have some information available about the kind of variants and the examples.  To shortcut

some questions now, below I post a control script that I use to produce the snippets, and you might want to refer to the

EQ modifications that were use for the several recordings where the mod was needed:

 

(Note my use of wildcards in the filenames -- please don't get confused about that -- mostly look at the decoding arg variations.)

In most cases, but not all, the recordings are 'listenable' without the EQ, but I am shooting for optimal results, and being as transparent as possible.

 

The commands are documented in the 'Start Using' document, but here is a short description:

--pvl=<freq>,dB  (1st order LF EQ with frequency, gain calculation)

 

The first argument is 'frequency', and I have sometimes see values as high as 2kHz, but only on certain ABBA  VoulezVous recordings.  Most often the values are 500Hz or below -- and then, usually 250Hz or below.

The 2nd argument is dB (strength) of the EQ, and is ALWAYS +-3dB and +-6dB.   +dB is a bass boost at the specified frequency. -dB is a bass cut.

Given that the optimum values ended up being +-3 or +-6dB, it appears that these were done on purpose to make the FA version of the recording sound best.

 

/music/Anne*/15/01* --fa --pvl=150,3 --pvl=75,3
/music/Anne*/15/10* --fa --pvl=150,3 --pvl=75,3
/music/Herb/H*/04* --fa
/music/Herb/H*/07* --fa
/music/Herb/S*/01* --coff=-2 --fw=classical --fa
/music/ABBA/j/ABBA/01* --fa --fw=classical
/music/ABBA/j/ABBA/02* --fa --fw=classical
/music/ABBA/j1/TheAlbum/02* --fa --pvl=150,3
/music/Beatles/The*/Abb*/03* --fa
/music/Elton/*2/01* --fa
/music/Elton/*1/04* --fa
/music/Elton/*1/05* --fa
/music/Petula/SinglesP1/*01* --fa
/music/Petula/SinglesP1/*05* --fa
/music/*FLAC/1975*/04* --fa
/music/*FLAC/1970*/08* --fa --pvl=500,-6
/music/*FLAC/1970*/06* --fa --pvl=500,-6
/music/*FLAC/1970*/04* --fa --pvl=500,-6
/music/*FLAC/1971*/05* --coff=0 --fa --pvl=250,-6
/music/*FLAC/1971*/06* --coff=0 --fa --pvl=250,-6
/music/*FLAC/1971*/09* --coff=0 --fa --pvl=250,-6
/music/Carly/old/02* --coff=-4 --fw=classical --fa
/music/Carly/old/05* --coff=-4 --fw=classical --fa
/music/Simon*/Disc*/Book*/10* --coff=-2 --fw=classical --fa --as=39
/music/Simon*/Disc*/Book*/11* --coff=-2 --fw=classical --fa --as=39
/music/Simon*/Dis*/Wed*/06* --coff=-2 --fw=classical --fa --as=39
/music/xmusic/*Mason*/07* --fa
/music/xmusic/Al*S*/1976*/05* --coff=-4 --fw=classical --fa
/music/xmusic/Al*S*/1976*/09* --coff=-4 --fw=classical --fa
/music/Linda/1977*/07* --fa   --pvl=250,-6
/music/Linda/orig*/07* --fa   --pvl=250,-6
/music/London/CD3/07* --coff=-4 --fw=classical --fa
/music/London/CD1/08* --coff=-4 --fw=classical --fa
/music/London/CD1/09* --coff=-4 --fw=classical --fa
/music/Nat/*ry/*1/15* --coffset=-4 --fa
/music/Nat/*ry/*1/11* --coffset=-4 --fa
/music/Nat/*ry/*1/01* --coffset=-4 --fa
/music/olivia/*Soul*/*01* --fa
/music/olivia/*2/*14* --fa
/music/olivia/*2/*23* --fa

 

[John Dyson]

As you all know, I have hearing troubles.   All of a sudden, I am hearing the need for an HF shelf of -6db at 9kHz.   Have the decodes been

too bright?   I cannot reliably tell, but I do know that IF the recordings do need a shelving filter, the value above is what is needed.  (There is an alternative

scheme also that takes advantage of the layer scheme -- by checking out distortion, I can choose between the schemes.)   It does seem like

the standard shelf is most correct...   (My hearing troubles are both high end and low end -- I lose one or the other, and then don't know which to choose --

usually making the wrong choice.)  Maybe my hearing is playing games again -- the decoder is so close to 'perfect', but I cannot do it without some

help.

What do you think?  I am trying to get the release out before having to disappear for a few days.

[PeterSt]

6 minutes ago, John Dyson said:

Have the decodes been

too bright?

 

As far as I can tell, on April 29 I reported about the "L" version of the time and explicitly called the highs OK. You can read back on it in your PM.

I don't know about highs in later versions because other things bothered me.

 

Peter