'FeralA' decoder -- free-to-use

[John Dyson]

This calibration matter is now fixed.  The actual bug was related as much about the LF pre/de-emphasis as the calibration itself.  Basically, if the LF pre/de-emphasis isn't correct, the LF leaking into the higher frequency bands will interfere incorrectly.  There is a precise LF pre/de-emphasis curve so that the interactions between the LF (and lower MF) regions interfere exactly correctly with the HF bands.

 

There is another issue about the HF phase de-scrambler.  The results of de-scrambling were not ideal, so the behavior was re-reviewed.   The general concept appeared to be correct, and still is correct.   Part of the anti-distortion didn't go far enough, therefore leaving the results somewhere between the continued 'compressed' sound and the proper descrambled sound -- probably worse than both.

 

Revisited, using tells from several recordings -- ONJ 'Take me home country roads' is probably one of the more egregious recordings.   Primarily using that recording for testing resulted in a very senstive subjective measurement.   After many, many iterations (probably about 30), ended up with the 'best' results of all tests.  The subjective result is that the 'take me home country roads' starts sounding more normal, and other recordings have better, more natural dynamics.  Al Stewarts stuff is esp good.  I cannot say that the sound is a huge amount better on some of the recordings that need the descrambling, but other recordings really, really benefit.

 

The most obvious downside for using the decoder:  had to decrease the output level by another 3dB.   The outputs were originally clipping once in a while with a 0dB gain through the decoder, then had to change to -3dB, now down to -6dB.   Actually, -6dB is probably more severe than needed, but it is best to use 'round' dB numbers, and the nomal numbers are like: 0.1875dB, 0.375dB, 0.75dB, 1.5dB, 3.0dB, 6.0dB, 9.0dB, 10.0dB ...   Sometimes numbers like 4.5dB are okay, but best to avoid -- keep everything really 'even'.

 

Starting another demos decoding session in a few minutes.  If all is well, I'll expect to do a release.  Otherwise, another set of fixes might be needed...

 

 

 

 

[John Dyson]

Earlier had discovered the details of the HF phase descrambling.   The original scrambling during FA encoding has an advantage of softening HF peaks while retaining a similar energy level.   Therefore it maintains a relatively similar sound while 'chopping' the peaks in an AC sense instead of a DC raw clipper.

 

So much for the artistic description above, my job is to undo the mess.  Since the descrambling was discovered, about 2 layers, now a 3rd layer has been revealed.   The energy is stored using an ingenius (of course) FA EQ design created all those decades ago.   It is analogous to a DC clipper, but is done in a semi-nonlinear AC domain.  This energy is packed into a pseudo-clipped signal, and needs to be recovered back to the same original shape as before FA encoding.  This is a tall, tall order.   The unpacking is so fragile that the DolbyA decoding had to be corrected further.

 

The attack/release shape is so critical that additional improvements to the DA decoder design was needed.   Once those improvements were made, the original limit to the descrambling quality had improved.   While doing the step-by-step reverse engineering before the DA improvement, there appeared to be a quality limit.   Like usual, I had initially thought that the result quality was limited because of spectral and resolution limits (22.050kHz, 16bits), but since I have been burnt by that kind of thinking, did further investigation.  Using some intuition, looked at the DA emulation, fixing all of the little 'shape changing' aspects of the decoder, ending up with a noticeable improvement.

 

Frankly, the original intent was to write the FA encoder, and use the foibles on the output of the FA encoder to feed back corrections to the DA decoder.   This  is only the 2nd or 3rd cases of DA decoder correction that has been possible so far.   However, even the DA decoder as improved about 6mos ago has been further improved.   The attack waveform was a little too fast because one of the internal control domain feedback loops wasn't tight enough.  (There was an excess delay in the loop, thereby making the attack push too hard, and overshooting a little too much.)  It was very much a 'chicken or egg' type bug that was fixed.

 

In a few minutes, gonna try again, again for a release.

 

 

 

 

 

 

[John Dyson]

Been testing/verifying/making minor corrects/removing-adding anti-distortion EQ methods, etc.

There is an hour or so delay, but well worth it.  I am hoping for a release in the morning (USA EST time), but as you know, my estimates are very unreliable.

This upcoming release will contain the most profound improvements that have been added to the decoder, other than bugfixes that had brought it up to 'tolerability'.  More of my claims about 'stereo image' will manifest in this new release.  Most of this comes from the more complete HF descrambling.   On the singular test recording that I have reviewed so far, given the same general volume level,. the signal peaks are at least 1dB greater, with the HF dynamics being 'majorly' stronger.

After doing the comparison between 'raw', direct from CD or digital source vs. decoded, it is amazing how much of the recording is 'smushed' by the FA encoding.   These new results are even bigger than my original thinking on this matter.   Don't expect a 100% improvement over previous good decoder releases, but on an arbitrary 'gut feel' scale, perhaps 33% improvement in dynamics alone.  (So much for mixing objective and subjective :-)).

 

The biggest hold-up right now is 2nd guessing the needed EQ after the HF 'unfolding' (hate that term.) or more accurately, descrambling.  I think that we have the correct answer, but need to do more verification.

 

Gonna test test test test before wasting anyone else's time on this...

 

 

 

 

[John Dyson]

During the test decode demos for V6.0.5S, got some interesting results.   This will require careful review, some changes, definitely not reworks though.

 

Summing up the comment below (saving reading for most people) -- after adding the *really good* HF descrambling, there is still likely an HF EQ error.   Adding the descrambling required some EQ changes, and like so very often, I might have botched the last EQ step.   The results are very close to being ready, but doing this next release is going to *try* to err on the side of caution.   There are some other, non audio things that might need to distract until Friday, but will work as much as possible on getting the release out -- I really want to do the release before Friday evening, EST US time.  It would be really nice to make the decoder available before this coming weekend for people to 'play' with.

 

======

 

The general sound is mostly of a little less 'brightness' on the decoded version, and there are some major surprises, true surprises of beautiful quality (IMO).  On the other hand,  notable on lack of 'brightness' is the sound of the Elton John recordings.   They don't have that  serious brightness that either the original raw recording or previous decoder output.  The results sound very good, but not the expected character.  Also, when listening to most of the demos, the new line of decoders seem to have fewer lingering FA effects.    The slightly weak HF tells me that the EQ choices after the addition of the descrambling might not be correct.   In earlier tests today, 'normal' recordings like the classical demos or Linda Ronstadts torch songs (Round Midnight) sound reasonably normal (ended up too mellow), but some of the pop recordings have lost some of their desired intensity.  This might mean that one of the EQ steps might be wrong -- the normal EQ methods can have strange/unexpected effects.

 

The general sound is VERY good and VERY pleasant, but perhaps missing some highs in the 6k+ range -- giving a sound that is a bit too mellow.

The changes will ALWAYS follow the 'rules', but just chose the wrong 'blocks' on this non-release.

 

It will definitely require another redo -- we are at '5S' right now, would like to release before '5Z'!!!

 

 

 

 

[John Dyson]

Here is a bit of status, and a demo showing the phase descrambler output.

As a bonus, a foggy version of the phase descrambler output.  ('Fog' isn't just fog, but a confusion factor in the sound.)

We are also in good shape for Friday early evening (will try for my afternoon for EU people.)

 

Now -- please remember, this is a work in progress, but the descrambler IS aligned.  (Yes, there is an alignment process needed to match the original scrambling.)  Also, when doing the phase descrambling, the pre-emphasis/de-emphasis MUST be very accurate, or nothing works right.  There are still some questions about some of the post-decoding EQ though -- so if you hear impairments, the HF post decoding EQ might still have some minor troubles.  Also, even though the descrambling is aligned, there might be still be too many/too few layers in the process.  Choosing the number of layers of complexity requires more careful listening.   After the URLS for the recordings (I hate for AS to spend the space), I'll make some comments about the phase descrambling stuff.

 

This is one of the test case snippets, and has an especially egregious problem. (ONJ, Take Me Home Country Roads, at 48sec.)  THIS IS NOT CHERRY PICKED!!! Much prettier examples can be shown, but that is not the purpose of this demo.  This DOES generally sound pretty good though.

 

RAW original:

https://www.dropbox.com/s/c0xjz3533b9anxk/onjphdRAW.flac?dl=0

 

DECODED with phase descrambling (sorry, cannot easily be disabled, all mixed in with post-decoding EQ):

https://www.dropbox.com/s/l77w0ruuxblzlpc/onjphdPHDES.flac?dl=0

 

DECODED with phase descrambling, fog (confusion) left in:

https://www.dropbox.com/s/q3urrrn6j0alh4g/onjphdPHDES-FOG.flac?dl=0

 

About phase descrambling...   Without the descrambling, the result is just a little better than the RAW copy.  It is the descrambling that brings everything up to a more normal sound.  When finally figuring out the descrambling scheme, it sure looks like a video compressor/expander that would be used on a VCR.  That design technique would come from someone with a video background.   This points even more towards R Dolby doing the FA design.  (R Dolby was on the team that designed the first Quadruplex VTR at Ampex.)

 

Again, this is just a 'status report' and to let you know about the 'phase descrambler' that I keep mentioning, and also another example of 'fog' (which isn't just 'fog').   One note of import when using the decoder (soon!!!), I found that the anti-distortion scheme can be a very blunt object.  I now have some existence proof that it goes FAR FAR FAR beyond removing the fog from the current processing pass.   It appears to remove ALL sidebands that it is tuned to attack.  When running the anti-distortion beyond a useful level, it produces some very interesting (I really mean interesting) effects in the sound.   From now on, the best decoding level is '--xpp=max'.   Any setting above '--xpp' will remove significant parts of the recording.  It is amazing at removing everything except the vocal!!!  Then, even with the vocal, something is lost.  This means that the only sidebands that can be removed must be in the region of fast attack/release speed. (above 200-1kHz).  Frequencies below that do untold evils to the recording, and IT CANNOT EASILY BE DETECTED!!!   I didn't realize this till today.

 

 

 

 

 

[John Dyson]

A new version of the ONJ snippets with the phase descrambling have been added to the small repo.

Instead of giving individual pointers, here is the directory that contains the original and -new versions.

It seems like the '-new' version might be better, but making the correct choice out of several is really tricky.

 

These examples and their differences should show how difficult it can be to choose, especially, the output EQ.   The phase descrambling isn't very difficult, because there are 'tells' if the phase descrambling is incorrect -- and there is a 'pop' when the setting is correct.

 

There are a *few* weak tells for the output EQ, and actually the output of the phase descrambler helps a lot.  However, there are still perhaps 2 or 3 choices which mix well with the phase descrambling output, and I am INCREDIBLY poor at making the most artful, correct, proper sounding choice.

 

It does seem like the '-new' versions are a little closer to what is found on a commercial recording, and just might be correct.  There was a slight amount of development between the original demos pointed to above and the '-new' versions.  This is all in an active state of development.  Again, BOTH sets of versions reside in the repo below:

 

https://www.dropbox.com/sh/imuvtb7zj4addrl/AAC9PkAucbTrK2khDu84izcLa?dl=0

[John Dyson]

Of course, the final version of the descrambler will be in the release...   Got it working very well, but have a lot more to do.  Found a fundamental mistake in the DA decoder, and must change some settings that might take a lot of time!!!  It will fix some problems in the descrambler -- been working for a few days on it.

 

The descrambler appears to have been a major part of the decoder, always missing until now!!!

 

 

 

 

[John Dyson]

The Friday release will have to be a preliminary, but functional release.   Filling out the whole decoder with the descrambler is very tedious.  It ended having lots of tentacles throughout the entire FA side of the decoder.   That, plus the DA decoder has had a *major* accuracy upgrade.   This was needed because of the extreme requirements of the descrambler.  Lots of changes, and a bit tricky to get everything working together correctly.

 

Remember the spiel about the passive anti-distortion?   Well, I was onto something -- and it is actually about the phase descrambler, base frequency sequence for the phase descrambler is the same as the magic anti-distortion numbers.  At this point, which one is the driving function for the other?  I don't know.   Many many parameters wrt 'anti-distortion' had to be changed, including my trusty 250,125,25,5 sequence, frustratingly -- it actually starts at 221.5 instead, and the factor isn't quite 5, but instead (2.215 * 2.215).   What about this '221.5'?   Take a look at the chart for 1% resistors.   A full review for aligning the descrambler by listening, just ended up with 221.5...  Then....  That number sounded very familiar.  (The 1% resistor values.)

 

Some of the more dense recordings are lining up more nicely, but the descrambler integrates into the post decoding HF EQ.   The descrambler really messes with the EQ horribly.  It is really a VERY cool scheme (as mentioned before reminds of a video design.)   Instead of doing a resistive nonlinear thing, it does a capacitive nonlinear thing of sorts.  The advantage of the capacitive nonlinear scheme, and the multi-layer approach to the descrambler is that it is not very dependent on signal level.  A resistive approach with resistive nonlinear devices (e.g. diodes), the scheme would be very signal-level-dependent.  *The difficult problem -- HOW TO DEFINE FLAT?"   The descrambler has NO static frequency response.  Tweaking by 'sounds good' really worries me.   There are some design patterns that have become obvious, so these are being tried for EQ correctness.

 

The dynamics seem pretty close to correct, but the EQ is still a bit harsh on complex material.  It is definitely not quite perfect, and given my schedule till Friday, there will only be a few hours to work on this release.   Even though the release will be preliminary, many aspects of the results expected on Friday will still likely be better than ever.  The ONJ example that I gave before it tending to be quite a bit better, but the decoder isn't really stable enough for a demo to show much more than before.   That is, any demos will not be close to the attempted Friday release at all.

 

A simple first pass of the phase descrambler was EASY.   Trying to chase down the entire design is MUCH MORE DIFFICULT than imagined.  IMO, it isn't sufficent to do a superficial implementation -- it must be as complete as possible!!!

 

 

[John Dyson]

Running first pass of full set of demos.   In a lot of sections, the decoder is almost a full redo, including a more accurate DA decoder.

 

The DA decoder had some problems with the attack time.   Most of the time not a big problem, but could make some decodes a little too intense.  The attack time is now probably a little better than a true DolbyA.   (DolbyAs are in a full loop feedback mode when decoding, that is where most of the errors happen.)  Since the DA decoder is not 'audio feedback', and in fact uses very little feedback for its control, only controlling the attack time and the speed of the release time (based on gain.)   Makes it more complex to do correctly.  There is a patent where there is a full control feedback, where the input bandpass filters gain is tweaked in feedback mode, but that really seemed to be frought with a few problems, most importantly it was patented.   I did a big patent search for the kind of method used, and the DA decoder doesn't work like anything that I could find.   It is even different than R Dolbys compressor patents, one reason, because there is no compressor.

 

Embarassingly, one of the DA decoders fixes was that the gain control calculations missed a full wave rectifier, which is one of the attack time problems.   Since CPUs that the decoder was inteded to use, and being lazy -- the calculation for the full wave rectifier is:   value = sqrt(input * input).   There are faster ways of doing it, but this doesn't need any bit tricks or conditionals.

Listening to the demos right now, and they are ending up sounding pretty good.  There are some more things to do, and will have to start the final version sometime tomorrow. It can still be screwed up, that is why the demos need reviewing.

 

 

[John Dyson]

Just had a chance for ACID test on the DA part of the decoder....

I was VERY surprised.

 

ABBA Gold and More ABBA Gold 1992/1993 are both DolbyA, but a little bit of EQ at 9kHz before decoding.  After decoding, it sounds best with a fast rolloff at 18kHz.  The result is NOT FA.

 

In DA mode, use --tone=-18.075   Use 9kHz,-3dB,Q=1.414 for ABBA Gold, 9kHz,-6dB,Q=1.414 BEFORE decoding.

Sounds great -- a LOT of dynamics.  Bright.  Can hear the 'twang' clearly (not loud) at 10seconds into SOS.

 

 

[John Dyson]

Demos looking good as-is right now...  We might be able to use the build that I already have.

Caveat -- it wont' be a 'final' kind of release, but I doubt there are many flaws.

 

Be ready for a dB, perhaps 2dB or so more dynamics, and it generally sounds natural.

 

Also, as you know, I am variably deaf to different regions of audio spectrum.  It doesn't sound bad to me on an A/B, but if I cannot hear super highs, it might be 'hot'.

 

 

 

 

[John Dyson]

The V6.0.5W decoder is uploaded, but after reviewing and thinking, I believe that the phase descrambler is slightly maladjusted.  The demos are NOT uploaded, and I don't think that I want to complete them since 'X' will be coming out in about 6Hrs.  Cannot distract because of the work needed.

 

  It is *really* tricky to find the correct EQ combination, because the descrambler does so much to the signal.  It is BEAUTIFUL to listen to, but remember I have HF hearing problems at times.  The descrambler EQ is maladjusted in the sense that the HF isn't as nicely  'wrapped up' as I want.   What I mean, is that it has very nice, slightly sparkling highs, but NOT with a 'tilt'.  So, it is pretty, but the descrambler output is a little too intense.  Oddly, material with loud ticks like some Dire Straits doesn't become overwhelming, but it is just a little too bright.

 

Since the descrambler isn't perfect yet, and after thinking and listening, I know EXACTLY what to do.   Intrepid people might want to try the V6.0.5W as it is VERY nice sounding -- again, not perfect as I plan.  So, in about +6Hrs, there should be a full V6.0.5X.   I really want to deliver it before the weekend, and I think that we'll be happy in a few hours.

 

PLEASE do not expect the constrained sound of the old version...  This is not just a multi-layer expander, but something much better in addition to the precision expanders.  The old version was better than FA (or at least the most recent internal version), but this new one will truly be pleasing, and I'd hate to screw it up with this very minor problem.  Earlier today, my hearing fooled me again, and I missed a slight EQ at 24kHz which has an phase cleanup effect down to abou 12kHz.  ITS GOTTA BE FIXED!!!

 

Sorry for the delay, but I truly want you'all to have something as good as I can do at this point.  Even the current 'W' is at the approx 6th or 7th attempt.   We'll get the 'X' release in one attempt -- no problem!!!

 

John

 

[John Dyson]

V6.0.5X decoder is ready.  However, there is a caveat.

 

Caveat:  I am half deaf right now, randomly losing my HF and LF and both sometimes.   If the HF is too strong or the LF is too strong, PLEASE tell me.   There are some 'switches' inside the code where IT IS LEGAL, SIMPLE and FITS INTO THE FRAMEWORK to change the LF or HF (above 3kHz) by 1.5dB or 3dB.   There are valid EQ patterns, normally I follow the 'rules' very religiously.  IF there is a problem, it can be fixed with new demos in under 6Hrs.  SORRY about this, but I am going coo-coo on this.   The delay was caused by my hearing, and all I need is feedback on which way to go!!!  THIS SHIFT IN EQ HAS NO BEARING ON THE IMPROVEMENTS, I carefully listen to the HF/LF alone and try to listen to the whole thing together.   SOMETIMES, I can hear normally, but not this afternoon.  Earlier, it sounded correct, the bass seems to be getting stronger and stronger -- all I ask is -- HELP!!!   Thanks for any kind soul who can give me an evaluation of the response balance.

 

When there is a perceived flat response balance, the results are BEAUTIFUL -- REALLY!!!

PS:  the response measures that the HF is abt 2dB higher than LF, but with the phase descrambler, it is hard to interpret the measurement.  I think that the measurement is catching the peaks -- so I interpreted as if there is a 2dB higher very short term average (almost peak) measurement to match the LF.

 

Caveat finished -- here is the location of snippets and decoder.

 

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

The decoder itself is in the obvious sub directory, and look for V6.0.5X in the name.

There are EQed versions (for sibilance reduction) and NOEQ versions.   NO EQ  versions really sounds smooth.

 

 

Super important usage about the dynamics!!!

The decoder produces *peak* dynamics about 2 to 2.6dB higher than the old versions of the decoder and the FA original. Turn down the volume until you get used to the dynamics. THE DECODER SOUNDS *VERY* different. PS: the dynamics sound GOOD.

 

Important special notes:

Updated DA (decodes Dolby A type material). Much better, more compliant attack time. Maybe better, more clean than the HW units!!!  That is *before* the anti-distortion stuff.

 

Phase descrambler for frequencies above 3kHz (actually reaches a little below 3kHz):

Makes a big difference, maybe more than the dynamics piece itself. It straightens up the high frequencies. The highs are spread during FA encoding. It compresses the peaks in a semi nonlinear way. The decoder removes that scrambling.

 

Still a little worried about sibilance on SOME recordings:

There are two possible solutions, depending where the problem sits. If the problem is in the recording, then a few state variable notches might be in order. Dynamically measure the ‘sibilance’ frequency region, looking for the right sibilance energy, and compress it down. Most of the time, this filter wont’ do anything.

 

If the problem is in the decoder, it is most definitely in the pre/de-emphasis. The pre/de-emphasis is a very tricky beast, and changing it requires great care and a lot of testing. However, if there is a problem where the sibilance is expanded too much, then there are some possible modifications, but the mods can have severe side-defects. It seems like the pre/de-emphasis is now correct, but dynamics processing expansion is a little less important considering the phase descrambling effectively expands the HF in the signal also.

 

 

 

 

 

 

 

 

[fas42]

John, you have mentioned ABBA's Super Trouper  several times, and I thought I would give you some immediate feedback - I presume 01 - Super Trouper-RAW-SNIP.flac is in fact the original track, yes?

 

Unfortunately, downloaded, the RAW is quite superior; it has lilt, and makes me enjoy listening; the V6.0.5X+5-0, DEC has a 'digital' quality to it, and does not engage, at all.

[John Dyson]

1 hour ago, fas42 said:

John, you have mentioned ABBA's Super Trouper  several times, and I thought I would give you some immediate feedback - I presume 01 - Super Trouper-RAW-SNIP.flac is in fact the original track, yes?

 

Unfortunately, downloaded, the RAW is quite superior; it has lilt, and makes me enjoy listening; the V6.0.5X+5-0, DEC has a 'digital' quality to it, and does not engage, at all.

Early, when they started using the FA process, the 'powers that be' did do a comparison between FA and the true raw recordings, many did prefer the FA sound instead of the original non-FA recordings.  It is *amazing* to me, but that is the way that it is.   That is why I strongly suggest that people listen to the version that they prefer.  I prefer the original, pre-1980 vinyl sound and that is why the project started.  It is only in this release that it is coming close -- except I cannot do the EQ reliably.

 

* I'd suspect in ABBA Gold, which is definitely two passes of DolbyA with EQ, some people will *definitely* prefer the version that is the distributed FA copy BEFORE decoding with DolbyA (and it is a REALLY, REALLY good decoder.)   I prefer the after-decoding version, not anything to do with which decoder I uses (real HW or my SW toy.)   I detest the compresse sound on ABBA Gold, and was the original reason why the decoder project started!!!

 

As I have been trying to find out, there is a  possibility of a need for a different EQ. Is the recording 'tinny' sounding, or does it have/need more or less bass, more or less treble? (the balance can be fixed to be perfect in about 6Hrs, but I cannot detect what needs to be done.)  It is also possible that the result is too edgy (because the descrambling has layers,) and I estimated the correct choice -- but if there are too many layers, the sound can be edgy. I cannot tell because my hearing changes all over the place.  Believe it or not, right after the decoding the demos, I heard TOO MUCH BASS in the 'X' version.  Yet, when doing original testing/setting up this version of the response seemed balanced and also,  I just followed the obvious EQ sequence in the code.  Now, till 10 min ago,as sitting back in the position that I sleep, I was hearing a tinny sound, I am sitting up, now I am hearing more and more bass.  It is so frustrating!

 

There is a 3dB knob for HF and a 3dB knob for bass -- with the possibility of 1.5dB also being valid.  I know that given the correct answer, and I make the change, the balance cannot be anything but perfect. It is trivial to fix, esp if you hear too much treble.  That is easier to fix than the bass is.

 

Already, I got a big-time YEAH!!! from one person, and a NAY from you.  (thats okay -- everyone will have a different opinion because we all hear differently.)   I bypassed the testers this time, but the first feedback was from one of the testers. I need just one question answered from a few more people -- the question in bold.  Even IF you don't like the sound, how is the BASS/TREBLE balance?

 

Even if someone can answer the question about BASS/TREBLE, I have to figure out which one is in error -- because the upper midrange is the zero point.  So far, I got feedback that the sound is enjoyable (from the standpoint of balance), but I want to fix it ASAP,  people give feedback in their own time.  I'd hate for those who like the decoded sound end up with an EQ problem.

 

PS:  I do trust the tester who gave feedback, but each person hears differently!!!  I'd like BASS/TREBLE feedback from a few more people -- just to let me relax for a while!!!

 

 

 

 

 

 

 

 

 

 

[fas42]

Quick answer, John - just had another listen; to my ears, the treble level seems fine, is in accord with the RAW. But I would only lead you down the wrong path, commenting on bass - the speakers just don't do it anywhere low enough.

[John Dyson]

23 minutes ago, fas42 said:

Quick answer, John - just had another listen; to my ears, the treble level seems fine, is in accord with the RAW. But I would only lead you down the wrong path, commenting on bass - the speakers just don't do it anywhere low enough.

THANKS SO MUCH!!!

Remember, I respect your opinion about not preferring the decoded version.   That opinion comes from each person.

However, those who do prefer the decoder are VERY VERY welcome to help me adjust, just as those who don't like it can help me understand why not.  I know that there are people who do not like it, but I'd love to learn the particulars.   @fas42 Has given some description about how he feels -- and more input and a sense of the perception is still helpful.

 

* I'll be producing a 'Y' version that might be more tolerable for those who don't like the 'digital' like sound.   It is a known mechanism in the decoder, and if it is what I think it is -- EASY TO CHANGE.   If the 'Y' version is preferred -- then we'll go with that one instead.

 

Maybe the decoder does need some changes, but I have *truly* followed the design patterns, it is just that there are thousands of particulars that must be correct for the decoder to sound as well as it *might* be able to sound.   For example, the descrambler is very sensitive to is base frequency and the ratios for each step/layer.  Any change from exactly 221.5 Hz doesn't sound right!!  There are at least hundreds of these basic architectural items, and many many variations of each one!!!

 

 

Even though I did get good feedback from those who like the decoder, I must consider input from those who don't like it.   It is possible that the comment about a clinical/digital kind of sound does come from a known  cause, it is just that it seemed better to me.   Making a change still follows the design guidelines.  I will produce a 'Y' version with a mellower sound that still follows the 'rules'.  Lets' see how people like it instead?   It will only take a short time to produce a version for  comparison.  I am assuming that the response balance is good enough for now.

 

[John Dyson]

As mentioned in the previous forum message, I'll be producing a 'Y' version with a less edgy sound.

It will be ready in about 6Hrs.  It is a *known* mechanism as noted before.   I believe that if it is in error, it is just that I made the wrong choice.   Thems the breaks!!!
 

PS:  the upcoming V6.0.5Y version will actually have a *wider* response, but be less intense.   Counterintuitive, eh?

 

[John Dyson]

Found some optimizations/major quality improvements far above the harshness.

It takes time for testing, but I am hoping to get this out very soon today -- maybe 9Hrs.   Trying VERY hard, this is difficult stuff!!!

 

 

[John Dyson]

Just started decoding the demos...

The changes include --

 

* one of the most important goals -- get rid of the edginess.   Do you know how you get edginess?   Actually add a -3dB rolloff above the audio range.   Some Q values cause a peak in the upper audio.   The new version is more carefully crafted and doesn't really have a peak.

 

A re-sequenced phase descrambler.   This new version descrambles better and deeper.   The previous version was a little clumsy, and the new one is a rethink on the numbers.   The structure is exactly the same as before.

 

With the re-sequenced phase descrambler, that has a strong effect on the HF output EQ.   I was very careful with the update, and won't release until I am sure (as well as possible) that the decoder sounds like what I want it do.

 

Using @fas42's terminology I think that we got rid of the 'digital sound' )the intense edginess mentioned above), if I understand what was intended -- it was VERY valid criticism.   Blame everything on my hearing, right? 😐(constructive criticism is appreciated very much , even if there can sometimes be terminology issues.) 

 

Changed the pre-emphasis/de-emphasis to be a little more conservative on the sibilance.   Frankly, I am not sure which is correct, the new pre/de or the old.   All I know is that with the new phase descrambler and the new pre-emphasis/de-emphasis, the sibilance problem is somewhat mitigated.

 

The frequency response is actually wider...  Even though the decoder sounds less 'sharp' sounding ("digital sound"), it actually has a better response curve (I hope.)  The EQ looks correct, and double checked with whatever measurement tools available.

 

When the decodes have passed my review (considering the impairments in the previous version), then the V6.0.5Y will be available ASAP.

 

It took an intensive effort, but hopefully,  I have understood the criticism.

Admittedly, I do sometimes respond with frustration, but that is ONLY because so much work had been done to create and finish the decoder.   Eventually, it is so important to review the critique though.  Both @fas42 and another AS person have given very quick feedback.   It is so good that the feedback came quickly, so the decoder could be corrected for other people to use.  The kind & constructive criticism is SO VERY APPRECIATED. 😀

 

 

 

[John Dyson]

V6.0.5Y is ready.

Lots of stuff cleaned up -- NO 'harsh' sound.  Hopefully this fixes the 'digital' sound.

 

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

[John Dyson]

Very important:

The decoder doesn't just try to 'improve', but REALLY sounds dead-on correct.  It is very different -- not harsh, but is dead silent with the music popping forward.  No sense of distortion caused by the decoder itself.  The architecture has been correct for about 1yr, (except for the phase descrambler), but FINALLY, it appears that all of the numbers have come together.  FINALLY.   Even the 'harsh' version was close to correct, but I had mistakenly allowed the harshness get into the EQ.

 

The big difference other than fixing-the-harshness is that the descrambler was re-programmed with slightly different parameters.

 

[John Dyson]

2 hours ago, John Dyson said:

Very important:

The decoder doesn't just try to 'improve', but REALLY sounds dead-on correct.  It is very different -- not harsh, but is dead silent with the music popping forward.  No sense of distortion caused by the decoder itself.  The architecture has been correct for about 1yr, (except for the phase descrambler), but FINALLY, it appears that all of the numbers have come together.  FINALLY.   Even the 'harsh' version was close to correct, but I had mistakenly allowed the harshness get into the EQ.

 

The big difference other than fixing-the-harshness is that the descrambler was re-programmed with slightly different parameters.

 

Thanks for the kind feedback.   I did find a minor problem.   It is about modulation of the sound because the passive anti-distortion being set incorrectly.   Basically, each EQ step needs to have the anti-distortion or not.   Orignally, on this series of releases, the sound was better with the anti-distortion on all EQ, with different parameters for each frequency range.

 

Unfortunately,I made bad choice above 3kHz (non-inclusive.)   With the sound being a little more normal (to me), I could hear the minor defect.  At the higher freqs, it appears that the anti-distortion should NOT be used.   The effect is a subtle rough sound instead of a pure sound.   The freq balance is really good (I hope), but the anti-distortion mistake is a subtle (very subtle) error that leaves the signal with a very subtle mistake.   I don't know if most people can hear the problem because my hearing IS almost impossibly good at detecting distortions even though I have absolutely no sense while trying to find a freq response balance.

 

If you hear the frequency response balance being relatively flat then the decoder should be usable for normal day-to-day (it should be almost perfectly flat, but I'll always double-check).  I am obscessive about perfection, even though it seems like I put out a lot of cr*p.  In a way, I have been -- but the complexity is crazy.

 

Of course, I'll be listening to criticisms, while also making the correction.   In approx 1day or 2days there will be a hopefully wrap-up release.  It seems like even the most simple change takes about 1day, so I wont promise quick turn around. HOPEFULLY, soon, we can work on source clean-up, a bit better docs, and improve usability.  Maybe even figure out how to use setup for Windows (any thing on Windows is a tall order for me.)

 

Things must be going MUCH slower for a few days, but the project will move forward.

 

 

 

[John Dyson]

Sadly, didn't hit the 30minute timeout on the previous message, but important to add on this info:

 

It is best NOT to use quality modes above --xpp or --xpp=max.   If you want better quality, then use '--dp.'   After careful listening, at highest 'quality' levels, 5 passes of the anti-fog can be very destructive.  One pass of the highest quality modes is helpful for DA mode, however FA produces very interesting results at '--xpppp' mode.  It appears to strip out too many sidebands.   The difference between '--xp' and '--xpppp' is that '--xpppp' reaches down to lower audio frequencies and probably starts stripping out some of the character in musical instruments.

 

Actually, 'fog' is removed farily well at --xp=max of --xpp=max, esp with --dp.  However, even just --xp does a pretty good job.   Fog is very, very evil on the FA decoder because everytime the gain changes at any of the 5 layers, then energy is injected into the signal.   On DA mode, there is only one layer of energy, and the anti-fog works great at almost any quality mode.  IMO, the FA decoder is unlistenable unless at least --fz is used, but that isn't nearly as effective as --xp, so --fz should be deprecated for anything but slower machines (like my i4770 4 core machine.)  Newer 8 core or the i12 series will certainly work at --xp with reasonable speed.   Even an advanced 4 core should be 'no slouch'.

 

Thanks again.   BTW, the correction mentioned above is working.  Testing/full verfication is now needed before the upcoming 'final fix' release.

 

 

[John Dyson]

Heads-up about the quality improvement:

 

Important improvement -- correcting the highs, keeping sibilance sane & keeping sibilance centered -- few times that it now shifts, less grainy.   Now, most of the time when there is a shift on the output of the decoder, it is likely damage in the original, RAW CD.

 

The phase descrambling(1),  HF output EQ(2), inter-layer HF EQ(3),  and HF pre/de-emphasis(4) all need to dance together.   They must all be mutually balanced.   Originally, it was a challenge just to create the 4 subsystems, and then to make them correct, sort-of.   The 'dance together' thing wasn't adequately considered any time before now, including in the previous 'V6.0.5Y' release.   Also, the phase descrambling is brand new.  It takes a while to learn how to use it.   Oddly, like all other major subsystems, there was no idea how to make each one, but also there was originally no idea that they were needed.

 

So far, more clean sound has been achieved, and better (generally more), sweeter HF.    The HF sound is infinitely more stable now, but the graininess is still at perhaps 1/4 of what it was before.   Interestingly, graininess doesn't just come from a non-linearity of some kind, but also could come from incorrect 1st order EQ.  1st order EQ is a tedious, irritating, overwhelming thing to reverse-engineer.   It is so much easier to work with 2nd order EQ, even though Q value is needed on 2nd order.  The graininess still needs some work.  The 4 subsystems are now 'dancing', but still stepping on toes too often.

 

The previous release is the first that reminds me more strongly of older (before 1980s) distributed recordings.  It hasn't 'made the grade' yet though.   Getting close -- no cigar yet. 

 Frustratingly, it has almost always sounded good to me and that is why constructive reviewers are so helpful.   It is so easy to be overly critical or claim impossibility for one reason or another.  The decoder will be finally proven to be 'possible', and working very well. 

 

Thank goodness for the discipline to 'follow the rules', otherwise it could never work correctly.

 

Thanks again for the help, and there should be a release tomorrow or day after.

Work will be done more leisurely for a while, but progress will be made, and the quality might even be more consistent.

 

ADD-ON:  5 mins after originally posting this message, the cause of the rest of the graininess has been found and corrected.

 

John