'FeralA' decoder -- free-to-use

[John Dyson]

The release is delayed a little, but NOT BECAUSE OF ANYTHING BAD...   Just the opposite -- got some feedback from an older version and want to make sure there are no regressions based on good feedback.  On just the normal 'demos', reasonably detailed checks comparing one release to another, can take an hour or so -- and must be careful that hearing perception doesn't drift.   Some feedback on recent versions is not just 'good' or just 'better', but very much the best feedback ever on multi-layer decoding.

 

Previously, 3-4yrs ago, there was some overly kind feedback, on 1 layer decoding. That feedback (and development) was based on misunderstood goals of what was needed.   I didn't realize that multiple layers was needed for ANY kind of accuracy.   The old 1 layer decoding is TRIVIAL compared to what is being done today.  The multi-layer interfacing and nearly the same settings for every recording has made the difficulty much much greater -- not just a little, but the difference between a 1 month and 5yr project.   Just the requirement for DA decoder accuracy and gain accuracy has been challenging.  (The slightest error on DA dynamics processing can make a very profound difference in the results.)  These issues alone are worthy of a full project, let alone the interfaces between decoding layers and the input/output pre/de-emphasis and EQ.

 

The stereo image thing does show improvement, but as you already know, I sometimes over-state things in excitement.   Even small improvements are becoming more and more difficult because the decoder has started zeroing-in to 'perfection' (whatever that means), and there have been so many small. incremental improvements now that it is difficult to stop and offer demos.   The improvement noted a few days ago does require good headphones or *excellent* speakers to notice, but it really is an improvement.  (When I mention *excellent* speakers, I am speaking of well time-compensated units.)   Correct and consistent time delays are now critical at the needed level of certain kinds of comparison.

 

I *almost* just offered some *status report* demos, but got some feedback that gives a baseline of performance, hoping to double check things.  Also, admittedly, it seems that the dynamics are  just a little 'soft' and believe that the sound should have slightly stronger dynamics.  This concern about dynamics is NOT about personal taste, but based on  comparisons with the few true recordings that can be used  for reference.   (The desire for stronger dynamics probably requires that the 6kHz EQ have a step step change in the pre/de-emphasis and the 9kHz pre/de-emphasis EQ be changed slightly to compensate.)   One simplifying issue is that the 3kHz pre/de-emphasis EQ appears to be set correctly, and any change causes disaster and rabbit-chasing.

 

In the 'test run' archive,  about 1/2 of all of the set of tests are completed (about 20,30 -- maybe 50albums), and would have stopped it if I heard ANY significant errors.   There *are* some residual defects about 'dynamics', some residual compression that will require some continued 'playing Tetris' on the building blocks.

 

Also, admittedly -- because of matters about reliability of hearing/perception -- a few days of distraction from the project have been needed.  Today, I  have been a poor judge on certain kinds of performance -- but what I can tell and based on feedback on recent one-off/requested comparisons has been good.   Once the judgement of the 'final tester' (me) is more reliable and sensitive enough to make more precise judgements, there will be at least a 'status report' release.   Private requests can probably elicit private examples tomorrow, but I don't want to botch the next public release.

 

This reticence is NOT at all about problems in the SW, JUST THE OPPOSITE.   The results are so close to the goals that on some recordings, it is difficult to distinguish between ancient vinyl and the decoded version -- except for the ticks and pops on the vinyl, and the more clean rendition of the material on the decoded version.

 

In a few hours, I might show a *fully EQed* and near-final test snippet, but ONLY if the perception by the  'final tester' is good enough.

 

 

[John Dyson]

I forgot to mention -- the recent stereo image improvement might be subtle, but the big difference is the stability and decrease in wobble.  Also the stereo field appears to have more consistency and fewer gaps in it.   It is almost  impossible to hear without a *really good* set of transducers, but it does exist.   Casual listening might NOT detect the improvement, and maintaining this spatially correct state is yet another set of criteria that is making the new release very difficult.

 

Real recordings (less-so FA) can really be unbelievably beautiful, and it might be difficult to accept that claim given problems with some of the earlier demos.  The goal at this point has moved forward and  is for nearer to perfection.  Even now, in more and more cases, there is a wonderful reversion from the FA state back to the original pre-FA state.   A *perfect* result where it is the same as the mix can NOT be attained, but something very close to the final product before FA encoding can and will be produced.  (some cases, audibly correct results, based on DIRECT comparisons have recently been attained.)

 

* Recently, there is more 'moving forward' than in previous weeks as set of guiding information sources has been re-organized to be more effective and a more consistent positive direction.   Constructive feedback has also been easier to utilize because the decoder is getting closer to correct (in fact, is producing almost perfect results in some cases).  There is a better information  'triage' than before, but also the quality of the user/critic provided help has greatly improved in the last week or two.  I have been learning to triage and discern aspects of critical feedback, and learning to better understand what each correspondent wishes to offer.   Basically, the  critical feedback is better understood/utilized and the resulting effective 'signal to noise' has recently gotten MUCH better.

 

If you compare the current internal test materials with ANY other public/private demo, there is already an improvement that goes beyond conscious perception (at least, in my experience.)  There are some improvements that I yet to have technically describe...  This is becoming very challenging, far far beyond my casual hearing/perception abilities and problems.

 

John

 

[John Dyson]

The upcoming V3.5.7A is in review by the early reviewers.   The major issue for me right now is that the bass/HF balance seems to drift around because of my hearing, but this upcoming release is DIRECTLY based on a release that a reviewer thought was pretty close to correct.

 

The new 'stereo image' is also included.   There were small holes in the stereo image on the earlier releases (those releases that sounded good at all.)   These holes were still smaller than FA has, but the new 'stereo image' code is interesting indeed.   A full analysis of the stereo image compression scheme would be educational, and really shows R Dolbys true genius.

 

Sorry about not making it this weekend, but the very earliest that the reviews might be complete might be the end of day today, but might be a day or so longer.   There might be about 1.5dB too much bass in the review version  (that number is easily modified), and that is actually a LOT of difference.

 

The decoded versions do match the non-FA, true, pure recordings fairly well.  In some cases the clarity seems a little better (probably because of the ancient vinyl.)

 

If I get a lot of confidence and if one of the reviewers gives me a bit of positive feedback, I just might upload to the public sites.

 

 

[John Dyson]

Good news -- just got some initial feedback, and it was good enough that there will be a public release.   Some very minor changes for usability.  This has little to do with EQ, but the highest quality modes were running just a little too slowly.   There is going to be a run of ALL demos, and because of the vast size of the catalog, will take at least 6-8Hrs (dozens and dozens of albums).   When all are complete, and I have done another review of the recordings, the release will be made available (and demos/snippets/personal requests/etc.)

 

* I really thank those who have helped/done reviews/etc.   As a matter of respect and privacy, I do not share the names without permission, but I truly thank those who have helped and given good, useful direction.   These thanks are directed towards those who have recently given helpful direction, but also those in the past who had honestly tried to help.

 

The upcoming release will be V3.5.7C.   The reviewer accepted test release was V3.5.7A, and I had made a pre-emptive bass-reduced version V3.5.7B, but was unneeded.

So, the V3.5.7C will be the same as V3.5.7A with a few minor improvements (very minor.)

 

From the feedback so far -- THIS IS A GOOD RELEASE.   If we do have a difference of opinion on some behaviors of the decoder, it should be almost trivial to offer user-accessible 'controls' for an alternate 'sound' or 'behavior'.

 

Don't expect the release until approx 24Hrs, I don't want to hurry this.   The 'V3.5.7A' snippets -- reviewed as acceptable, are available at the location below.  The V3.5.7A version is not the full quality released versions but are prelim.   These test versions were not done at the normal high quality demo modes and are at a reduced quality for quick run-off.   The proper, V3.5.7C version, along with decoder binary, private requests, etc...  will be coming in the 24Hr timeframe.

 

Early access review copy:

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

 

 

John

 

[John Dyson]

V3.5.7C....

 

Demos:

Snippet demos:

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

 

From what I have been told, this is NOT the release to 'poo-poo' as an artifact of my bad hearing.   Four sources of corrections have been used:  true pure recordings as reference, 'replacing wierd headphones', 'knowledge about my bad hearing', and have been getting constructive help with kind motives for success.

 

No updated manual yet, and there are SOME recordings that need a very mild post decoding EQ (e.g. reducing bass.)  Just use '--fa' for the decoding command.  NO arguments are needed for '--fa'.   ANY need for EQ is post decoding, and I am even getting fairly clean results from 'Dreamworld'  (the good decodes of 'Dreamworld' show that the internal pre-emphasis/de-emphasis are correct, or very close to correct.)

 

I'll announce the updated manual -- the decoder IS worth using.   NONE of the adjustment has been based on 'tweaking', but once the generally correct sound is found, the results are locked-in by the relatively coarse jumps by moving 'building blocks' around.  The earlier dependence on the bad hearing/bad headphones and sometimes very very questionable 'help' (source has since disappeared), had been the most recent set of problems.   The decoder, working really well, is still EXACTLY the same design as months ago.  The problems were ALL about some small aspects of EQ that I just could not do by myself. 

 

*THE RESULTS ARE REASONABLY (very reasonably) close match to early vinyl.

WE NOW HAVE TRUE, PURE recordings as reference -- questions about 'not liking the sound' are now just a matter of preference, not so much 'correctness'!!!   Certain kinds of minor errors still might happen, because in certain kinds of adjustments,  the 'coarse jumps' sometimes  difficult to hear.

 

The current reviewers pointed me in the correct direction for EQ...   Even then, there were some contributors early on who really tried to help also.  Most, even from the past, have helped by their positive vibes.   There was only ONE person with an apparent malevolent intent -- but that can happen ANYWHERE.   I still must thank Alex, he really did try to help.  Wish he was still here so I could show him some success.

 

The private requests are being slowly built up also, they will be appearing over the next few days.   There are some private requests that I hadn't bothered doing until now -- On V3.5.7A (V3.5.7C is a slight improvement), there has been enough really good good feedback that it is time to fully show the benefit of doing the 'restoration' process.

 

Snippet demos:

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

 

Decoder V3.5.7C:

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

 

 

[John Dyson]

Heads up -- I just got feedback on ABBA SOS -- it is IDENTICAL (plus or minus minor differences), to the original vinyl.  Don't expect the bright, sparkling and FAKE sound of FA recordings.   I am making  'LESS BASS' versions with special mastering (same decode exactly -- DIRECTLY from the original decode, just additional EQ) for those who NEED more FA style sound.   They will have 'lessbass' in the names.   You might notice that with less bass, the result is reasonably bright for non-FA.   As I have written before -- THE GOAL IS NOT MASTERING.   If one wants a different mastering, the decoder can do it for you.

 

The less bass versions (again, no treble boost) was modified by the following decoder --equalizer command:

  --pvdl=1k,-3 --pvdl=250,-3  --pvdl=75,3 --pvdl=25,3

A modified version of 'less bass':

--pvdl=1k,-3 --pvdl=250,-3  --pvdl=75,6

 

Some recordings might require additional mastering/re-mastering.  However, the temporal behavior is IDENTICAL to the old vinyl, and the only real difference is the EQ (if any.)   The old vinyl copies that I have for ABBA have been EQed to hell and back, so it is difficult to guess at what they really were.  The FA version is also very distorted WRT EQ.

 

 

 

 

[John Dyson]

This is the result of post decoding mastering, derived DIRECTLY from the decoder output as demoed on the demo sites.

My own vinyl rip from early/pre-digital vinyl was EQed to h*ll and back, so which is the correct version?   The direct decode is the correct verison,

but I created some versions created directly from the decoded copies.

 

Here are the EQed decodes for ABBA (with less bass, some variations.)  more variations are possible AND VALID:

 

https://www.dropbox.com/s/c3bax5ywgcyxpl4/02. ABBA - Mamma Mia-V3.5.7C-0-DEC-SNIP-lessbass.flac?dl=0

https://www.dropbox.com/s/l9gls58oqthr1d4/01. ABBA - SOS-V3.5.7C-0-DEC-SNIP-lessbass.flac?dl=0

https://www.dropbox.com/s/ahdob483n0qkvum/01. ABBA - SOS-V3.5.7C-0-DEC-SNIP-lessbassDIFFERENT.flac?dl=0

https://www.dropbox.com/s/fc46nzwq2wdjy8b/02. ABBA - Mamma Mia-V3.5.7C-0-DEC-SNIP-lessbassDIFFERENT.flac?dl=0

 

[John Dyson]

Got most of the ABBA catalog decoded, will be providing snippets soon.

Supertramp has been done for several days, but Supertramp snippets just don't make any sense.

 

Same with Brubeck.  Since it is a natural recording, with no real processing, it provides a good baseline about the performance of the decoder.   I might be able to find a song or two from Brubeck that a snippet might be useful.

 

A lot of recordings don't 'do much' until 40 to 60seconds from the beginning.  I dont' have time to trim the recordings for each snippet.

 

Because of the settings that I am using, the results are a little more bright than original vinyl, but are not as intense as the current style of processing.

 

I have been getting feedback that the bass is correct, but I believe that it might POSSIBLY be a little strong.   More feedback on that area might be helpful.   I cannot tell if there should be more bass, but I must say that the current amount of bass is the natural amount given the EQ progression in the decoding scheme.

 

For the high end....  The high end is all in the recording, but I believe that the bass might obscure it a little.   It is a difficult tradeoff to make -- I just don't know what to do about the bass/HF tradeoff.

 

[John Dyson]

Some ABBA stuff has been added to the public demos.

.

Because of a hearing disability, mastering is difficult for me.

However, these results just might be closer to the response balance

closer to what is normally expected for ABBA. 

 

Even if these decodes have flawed EQ, they DO show the extreme cleanness of the

decoder.   Since the decoder is so clean (better than physically possible in normal

HW techniques), that almost any sane EQ will not normally expose any distortion created by the

decoder itself.

 

ABBA Album snippets: https://www.dropbox.com/sh/tvovud45r6vdc2n/AABxZXp1gr0VReHrlTW-cKnqa?dl=0

 

Full set of snippet demos: https://www.dropbox.com/sh/i6jccfopoi93s05/AAAZYvdR5co3-d1OM7v0BxWja?dl=0

 

 

[John Dyson]

After all of this work, and some more testing, I still believe that the vocals are a little buried.

I think that others agree with it also, but otherwise the response balance appears to be reasonably good.

Since we have no specs, we have to base the processing on what sounds correct instead of knowing the correct answers to begin with.

 

There is a remedy for the buried vocals, but like anything, it is best not to 'just add it' when not sure that it is the best thing to do.

During the 5minutes after addding the very minor change, it seems like the vocals (e.g. Anne Murray's Shadows in the moonlight) make the vocals more pronounced and probably better balanced with the background.

 

When the MF->LF adaptor is changed slightly to afford the additional stages that bring the vocals more forward, and then the additional stages that make the correction for the modification, the vocals are definitely more forward without apparent negative consequences.

 

I WILL make some demos by later on today (probably under +12Hrs) and a release for personal testing.  As you probably already know, I cannot hear frequency response balance in an absolute sense, but I can tell that the general balance hasn't changed given the new mod - so the response balance  should still stay good.

 

Most likely, the release will be something like V3.5.8X, where 'X' will be a letter between 'A' to 'Z', most likely 'A' or 'B'.

Will announce when tested and ready.

 

 

 

[John Dyson]

As I have written, the vocals are too buried.   There are methods to bring vocals forward, butI don't like adding things unless really needed.

I have pulled the modification that was already in reserve, and added it.  (about 9 lines of code addition.)

The positive side effect is that the frustrating variation of post decoding bass EQ is almost totally gone.

(Actually, it does appaear to be entirely gone - makes using the decoder much easier again).

The most important positive effect is that vocals are much more forward without any other perceptable change.

 

As you probably realize, I have to keep all kinds of tools in reserve because the required tools aren't known ahead of time.

The new release -- looks like it will still be V3.5.8A or V3.5.8B, and it is looking like V3.5.8A.  (the change is so simple

and working so well that I don't think that internal releases will be needed.)

 

 

[John Dyson]

V3.5.8 is ready.

TOTALLY reworked EQ.   TOTALLY more clean sound.

Much less compressed sounding output.   Even though the sound is more smooth, there is actually a LOT less compression on the output.

 

That HF intensity should be gone, yet the HF should seem to be extended further.

Vocals more forward, different output matrix (basically different output mastering.)

 

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

 

(This took a long time to do because my hearing came back, and was really upset with the highs on earlier releases.   The fixes required studying and modifying the 'Lego'/'Tetris' block settings.)

 

 

[John Dyson]

 

Been working hard on trying to stabilize the final EQ -- my hearing has been playing tricks again and there are NO specifications or objective basis of comparisons available right now.  

 

This FA encoding is obviously a trade secret, and there are no legally available documents available  describing  the encoding method of consumer recordings.  There is NO WAY that access to documents containing trade-secrets would be acceptable for development decisions.  This project is totally clean WRT patents and trade secrets, and development has done EVERYTHING possible to avoid infringements.

 

It is amazing that progress on the decoder has gotten as far as it has -- the dynamics (general amount of apparent dynamic range, fade-ins, fade-outs) are IDENTICAL to whatever before-early-80s reference materials that are currently available.   The 'tells' that have allowed the building blocks in the decoder to mutually define each-others behavior DO NOT exist in the final output EQ stage.  This makes the final-EQ totally dependent on aural judgement of frequency response balance.

 

So, there will be further online releases and demo/tests available, but there will not be announcements of presumably fully functional releases until I can get some help with the HF EQ, or at least gain more confidence in my ability to judge the correct building block settings to use.   I will continue to announce notable changes and improvements as described further below, but after all of these attempts, it would be deceiving myself that I can trust my ability to judge frequency response balance.

 

* It is still VERY EASY for me to detect all kinds of distortions (e.g. the terrible dynamics distortions on consumer recordings), but my ability to judge response balance is totally screwed up.   Progress can still be made, but will be reserving my claims of full quality releases until I can get some trustworthy & reliable help on frequency response balance.

 

Ongoing DSP/Software design challenge:   some of the HF settings can-not be prejudged from an engineering standpoint because there was a lot of flexibility for the original designer.   There are several plausible kinds of settings, then within each kind, there are often several 'sub-settings'.   There is only one set of HF settings that is fully correct, but it requires good and stable hearing to compare with both FA and original recordings.   (The nasty sounding FA RAW recordings do have SOME correct amplitude information in them, but are difficult to judge because human hearing doesn't perfectly represent what the FA recordings actually have in them.)  Basically, FA (normal consumer recordings today) has a lot of unnatural kinds of signal in them.

Significant improvements the last few days:

Sometimes I have been disappointed in the transient characteristics of the highest frequencies.   The highest HF are in the result, but there was some kind of time delay causing some sloppy sibilance, almost as if there is a high frequency time-biased lisp of some kind. This HF delay has been fixed, and now the cases where there was sloppy sibilance are now corrected.  This 'error' was probably a result of non-linear phase HF filtering in the original recordings, perhaps the use of 2nd order analog filters before digital encoding.   When adding some correction for the 2nd order filters that MIGHT have been used, the the sibilance significantly cleans up and improves in quality.

 

Also, the lowest LF had been troublesome and after doing some very careful comparisons with ancient original recordings & corrections, the lowest LF processing NOW produces results much more similar to the pre-FA encoding sound.   Frankly, there was way too much lowest-LF than should have been, and older recordings had more of a 'resonant' kind of LF.   This 'resonant' LF is not really resonance, but results from how perception responds to certain kinds of LF rolloff.   The decoder problem appears to have been TOO FLAT LF rolloff in the 20-40Hz range.

 

Significant improvements are still being made, and the quality is much better than I would have ever expected.  However, this last issue of the last phase of frequency response balance is very difficult for individuals with hearing disabilities.

 

 

 

 

[John Dyson]

Yesterday, was doing searches DEEP into my archives, and found some ripped early Carpenters LPs.   Been bouncing between the Carpenters LPs, Anne Murray FA examples, and early ABBA LPs.   I know that the set isn't fully representative, but checking the archives for more interesting nuggets.

 

So far, found some serious errors in the previous bass beahvior (very serious omissons) and a few other minor things.

Progress is being further made.

 

I realize that the decoder has been much trickier than I had originally thought - but it IS getting better!!!

 

 

[John Dyson]

More REALLY REALLY good news...

After a lot of experimentation and reverse engineering, I found that the lower MF and LF ranges do NOT use normal EQ at all.   This is one reason for the improved, but still distorted stereo image on previous versions.   Also, the bass was somehow not 'full' enough.

 

The new LF/MF EQ fully uses the matrix EQ scheme.   Even the alternative EQ recordings (some old Carpenters, some ABBA, some Linda Ronstadt) use an alternate matrix based on the same scheme.   This resolves the incomplete or slightly dissatisfying EQ on the 'bass heavy' recordings also.   Just doing a rolloff at 75Hz and/or at 250Hz doesn't quite 'cut it'.   Interestingly, some of the matrix base frequencies are EXACTLY the same as the attempted straight EQ being done before, but the usage of those frequencies and the method of 'shelving' is quite different than anything that I am sure ANY of you or I have seen before.   This goes FAR FAR beyond the difference between 1st order and 2nd order EQ.

 

Preliminary feedback on very private one-on-one demos show a favorable response.   These responses are from people who are very willing to politely criticize the results.

 

Before the release (still a test release, but VERY interesting one) I still have to categorize the 'alternative' matrixes and formalize/organize the ad-hoc nature of the current research implementation of the matrix EQ.   These matrix schemes really do need to be documented.   Also, this EQ matrix has NOTHING to do with a stereo matrix scheme because all matrix operations are done on a single channel itself, and is an EQ matrixed signal wraped onto the signal itself.   (An ingenius way of scrambling the phase of the signal to increase average power levels on FA recordings -- that is, to make FA recordings more 'broadcaster friendly'.)

 

Also, I am doing some reviews to make sure that the HF EQ doesn't also use a variant of the matrix scheme.   It actually looks like it DOES NOT, but need to verify that for my own satisfaction.

 

If all goes well, I'll try for a Thur or Fri release.

 

John

 

[John Dyson]

Still working on the release.   Finding every little nit.

 

The clarity is showing to be outstanding, it is far far more detailed than most have ever heard WITHOUT false 'enhancement'.  Response balance still might be imperfect, but recent private tests have been very positive.  This new version has a seeming order of magnitude more detail than previous versions, but without false enhancements like often used 'elsewhere'.

 

Testing and review is very intense, and any nits are being corrected -- then restarting the tests with a new version increment.   Like always, every version ever created is available for comparison (well over 1000 versions, but only a few 100 are usable with the current command structure.)

 

Right now, the demos for V3.6.6A are being run.   I won't know the 'answer' for at least 6hrs, but if V3.6.6A passes, the release will be coming very soon thereafter.   There is a REAL chance that the version currently being tested will be the release.  If it fails, NO MATTER WHAT, the detected failing will be corrected, and testing restarted.

 

Massive amounts of phase correction/de-scrambling is being done on the new release.   Geesh, these consumer recordings are profoundly damaged relative to a real mix.   As implied above, the detail is almost perfect, yet no intentional HF imbalance.   Any response balance correction will NOT loose the extremely clear detail.   This improvement in detail and phase coherency has NOTHING to do with 'enhancement' .   The only intentional processing is for correcting the commonly used MISGUIDED processing.

 

Apparently, the FA (normal consumer recordings) are even more damaged than what I had previously believed.   Do you remember the 'Aphex' device that adds phase distortion?   The FA signal is damaged like that and even more so.   This phase scrambling apparently helps to give a part of the 3.5dB loudness advantage available for the normal consumer recordings.

 

 

[John Dyson]

Still working on the release.   The general sound will be identical to the FA RAW in many ways, and it is very close in the testing.   It can decode RAW FA to bee old pre-FA vinyl very very closely (better in the sense of immediate/present sound, but not with any extreme change in dynamics.)  Freq characteristics are also identical.   There are still very minor differences (inaccuracies) in the dynamics for the 6kHz and 12kHz region.   Matching the results requires repetitive A/B comparisons, and when moving a building block to soften the dynamics in one region can have effects in other areas.  'Tetris'.

 

I do understand the problems with the very subdued sound of certain versions of the 'decoder', but there had been problems with my expectations for the target sound character.   Apparently, even normal releases are somewhat more bright than a raw recording, and I had been force-fitting the recordings into what I remember the raw sound to be (which IS accurate.)  Unfortunately, my goals did  not match the sound of the original input to the FA encoding scheme.

 

The results are coming very very close to correct, and without an A/B or without being sensitive to the intermod characteristics of the raw FA (consumer) recordings, it might be difficult for many to chose the decoded VS FA as being the original.   My own impression is that the decoded sound is now similar to the FA, but with more clean sound in the mix of instruments and sometimes more prounced dynamics, esp on classical.

 

 

[John Dyson]

After lots of testing, double checking and small revisions -- the results now match the general sound of the FA raw versions, but without the hiss, intermod, etc.   It does seem (by my comparisons with otherwise provably unmodified recordings) that there is a general trend to give an artificial HF boost to recordings, probably at mastering time.   There is usually a simple EQ to bring recordings back to sane, but again -- by default, the next release decoder produce EXACTLY the same general FA freq response balance  without the FA distortions, hiss, etc.   THERE IS NO ATTEMPT OF REMOVING THE SHRILL FA SOUND.

 

I am doing some final testing, and generally wretching when listening to the raw FA versions especially.   I can understand the previous decoder outputs seeming to be a little dull sounding, but the recent deocder  outputs WERE close to the sound of pure recordings without the approx +3dB at 9kHz and +6dB at 18kHz boost that appears to be the norm nowadays.  (it ends up giving actually about a 3dB total boost at 9kHz and 6dB total boost at 18kHz after going through the filtering.)   Therefore, to come back close to what a real recording should sound directly from the mix - a small HF cut would be in order.   The current decoder produces the FA freq response balance result *DIRECTLY* and there is nothing but a direct target, and with very minimal EQ.

 

The next decoder should not disappoint those who like the FA balance, but don't like the 'grain' in the FA sound.   The tests have gone through about 75% of my library at this time, and will be doing more spot checks today before release.   For the next release, the basis of comparision is PURELY the FA original without attempt to correct back to a real sound character.  Both the simple --fa command and the seven layer '--fa=+7 or --f7' commands do produce the same shrill sound of the raw FA original.  (Geesh -- it isn't me who is hard of hearing!!!  I am *really* surprised that there havne't been complaints about hiss in consumer recordings.)   The shrillness is what I had been fighting. The HF EQ is now very very simplified from the previous decoder version, and only a slight pre-emphasis/de-emphasis needed for correct 6kHz range dynamics.

 

I'll be trying for a release tonight at +18Hrs or perhaps +27Hrs, because the testing and comparisons will be 100% complete and accurate.   Like usual, there has been no sub 3dB tweaking at all, and all building blocks are in 3dB or 6dB increments per the very likely original designers choices.

 

 

 

 

[John Dyson]

V4.0.0C is ready.   This new version has a response balance VERY similar to raw FA, but without the phase & dynamics distortion garbage.

 

V4.0.0C does NOT sound like previous decoder versions.   The goals had changed on this version, and there is no atttempt to correct mastering -- which I didn't realize that was what I was trying to do.  (I really wish someone in the industry would have helped!!!)

 

If there are still minor response issues, there are two simple places that can be changed  into simple 3 choice knobs.   Sometimes, I think there is too much bass, sometimes too much HF.   These results come as a result of back and forth/A&B tradeoffs.   Some choices are very tricky to make.   With the possible 3 way knobs and some feedback to me -- the SW change can easily be made.   Especially frustrating -- some recordings need a slightly different LF EQ mode!!!   Linda Ronstadt and early Carpenters both manifest that problem.

 

* possible deficiency that I really worry about is +-1 to +-3dB at maybe 20kHz.   it is easy (trivial) to fix by changing a building block frequency, but I have done as well as my hearing allows.  Likely, only ONE building block might need changing, perhaps 2 at most.   No matter, even if there is some error in the HF, but result should seem realtively clean to most people.

 

Demos: (Not all are fully populated at time of posting -- will be filled in over the next few hours.)

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

 

Decoder:

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

 

Private/individual decodes are being worked on, probably ready tomorrow some time.

 

Major changes:

FULL rework of BASS using a matrix EQ scheme (not a stereo matrix, but a non-topologically linear EQ sequence.)

Substantial rework of HF, using a different matrix EQ scheme, and also phase descrambling.

Further refinement of both BASS and HF inter-layer EQ.   The HF inter-layer EQ has more steps above audible range, but of course, needed for proper phase and gain at the highest audible frequencies.

 

The sound is essentially the SAME as the FA original, but without the grain, compression sounds, swishy highs, etc.

Differences between raw FA and decoded are all about the strength of the dynamics.  Less HF compression might disappoint some people on some recordings (the low level HF details remain low level as on the master.)  There is NO HF boost at low levels, therefore it sometimes seems like there is an actual loss of detail.  As discussed before, the detail is intact, but at lower leves.

 

There *still* might be slightly too much or too little of the highest HF.  It is trivial to change the EQ (it is a matter of choosing whether or not to use some 24kHz and 30kHz 1st order EQ...  YES, they do affect the highest audio frequencies.)   Almost all EQ is at 6dB increments, and that makes me feel very safe that there is no tweaking.  As usual, all frequencies are 'standard', with no tweaking of frequencies either.   All changes are based on using/not-using the building blocks and choosing things like a combo 9k/18k or just 9k EQ scheme. (A 6dB EQ gain increment does NOT necessarily map to a 6dB signal level difference at frequency.)

 

The previous problems were about it apparently being impossible to do simple EQ to get correct results.   Frankly, I kept trying to force fit a standard EQ sequence into a really strange phase/frequency response scheme. Out of extreme frustration, I did a full revisit of the EQ, realizing that it is IMPOSSIBLE to do the LF->MF EQ by a simple EQ sequence.   The EQ is a kind of skewed thing, but ends up sounding VERY VERY similar to the FA original.   Since most of my tradeoffs are made on 'tells' and timing errors, very often the frequency response suffered.

 

* When working on software for a long time, listening to the results, the 'accomodation' problem manifested -- just like it can manifest for other people also.   When doing comparisons with older decoders, I do feel a sense of embarassment, but cannot look backwards too much.

 

The HF phase-descrambling was originally called 'anti-distortion' scheme, but it wasn't quite complete at removing the phase distortion, but came close.   There is another step of phase scrambling correction just finalized today.  (The new layer of phase descrambling had been in place for several days, it is just that there were some bug fixes that finally made everything fit together.)

 

 

 

[John Dyson]

This release got good feedback so far.   There appears to be fewer eccentricities.

The only complaint so far is that the bass might be a little weak, and I understand that.

There will be a release tonight (already built - probably V4.0.1A) with the requested mods.

 

1)  The bass at 75Hz will be 3dB stronger (and that is A LOT.) - I have troubles comparing and judging LF.

2)  Q of final EQ at 150Hz will be moved from 0.8409 to 1.18.

3)  The highs at 15kHz to 20kHz will be about 1-2dB stronger.

 

Got good feedback on clean ness of sound, and sound qualities in general.

 

There were some decodes in the demos with bad settings, e.g. Olivias recordings were a bit middle bass heavy, but there is an explicit

setting to remediy that problem.  (There are two general bass submodes (--lfeq=110 and --lfeq=220) which tame the bass,

mostly even the strongest mastering for bass.   However, one can add in a '2' or '3' to trim the absolute lowest freqs  (e.g. --lfeq=112 or --lfeq=222)

and should be little needed.

[John Dyson]

Found the reason for the troubles with the bass...

Apparently, there is the goal of weak encryption in the signal.

Unless the 75Hz and 150Hz EQ is offset by the same amount of 50Hz EQ (approx +4.5dB), then the result is WAY WAY too much bass.

 

Again:  to DECREASE the amount of excessive bass, then INCREASING the amount of 50Hz makes it go away.

The scheme made the attempts at simple EQ totally futile!!!

 

There is nothing that I know of that can cause this effect other than a careful use of natural phase shift (unlikely) or an actual phase shift introduced to the signal

to make simple EQ almost impossible.   The EQ scheme is very intricate.

 

The new release is being reviewed by others right now.  Once I get feedback, and correct minor errors, you might REALLY be amazed.

 

John

 

[John Dyson]

V4.0.2F decoder is in pre-review. 

Will be announced in 24Hrs, giving the reviewers time to do a final check. (need an answer on the 6kHz matter as described below.)

 

1) strident sound gone.  (over-prioritizing the expansion for the highs, now corrected -- easier to get flat HF response.)

2) bass IS correct with new equalizer design (not just 1st order EQ method.)   Really cool.

3) The results are actually fairly pretty.  And very clean sounding...

4) Has been checked on every genre available to me.   The only really questionable result is on one ABBA cut -- just not sure if it is correct, because I don't happen to have a good pre-FA example of that.

5) Today, I just had about 15minutes of clear hearing, which gave me a chance to fix the HF bug that had been vexing me.  There still might be excessively strong HF, but if so, should be relatively mild.

 

With the smoother highs, there might have been a slight over-correction/error that is effectively -2dB at 6kHz to 9kHz range.   I am not sure -- there are two styles of 9kHz HF Equalizers, one is a two step 9kHz/18kHz scheme and the other is a one-step 9kHz scheme.   The sound is very similar, except the 9kHz scheme suppresses the 6khz to 9kHz range a bit more.

 

As an example, on Steely Dan's 'Peg', previously there was excessive conjestion in the vocal.   There was a tradeoff because all previous attempts to fix the conjestion made the Dire Straights (sp? recordings sound creepy and slimey.   I found the correct EQ after repeated listening/adjustement/listening/etc.   This new pre/de-emphasis EQ is actually super-simple and sounds the best after trying numerous building blocks.   Both 'Peg' and the Dire straights examples now sound clean.   Also, Anne Murry's recordings don't make Anne yell at you, and the sound is much more subdued (but still clear.)

 

I have done LOTS of cross checks to make sure that the response is nearly the same for 5 layers and 7 layers, which means that the inter-layer EQ is correct.   The input/output EQ has been a bit problematical because partially my hearing varies.   Also, the input EQ has fewer tells now because the decoder works so well and produces almost no dynamics distortion.   The only 'tell' that I have now is the HF expansion, which I just realized that the problem with the HF was NOT frequency response but expansion in the HF taking prioirity over the lower frequencies (it is a kind of cool scheme -- but doesn't change expansion ratio, even though it might sound like it.)

 

 

[John Dyson]

V4.0.5A is available, and well worth giving a try.

Been getting help from some people who are really good at describing the problems, and have the best intentions!!!

One person who frequents AS deserves a special thank you, and will do so publically with permission.

Just do some quick reviews of the demos.   Very little will disappoint.

 

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

 

There is a fantastic and stable stereo image on this version.

The HF might still be a little hot but the previous test version was also a bit hot, and this one has been toned down and also the structure of the attack (e.g. strings) are now more realistic without the overly harsh attack.

The BASS is good, likely relatively flat and not overwhelming.

EVERYTHING is decoding 'out of the box' except for the old Carpenters, Linda Ronstadt, but there is a switch that automates all of the change in EQ matrix: "--lfeq=alt0"

[John Dyson]

Heads up about a VERY VERY VERY subtle deficiency -- and might be the very last one.

The current V4.0.5A version is well worth trying even though this hyper minor deficiency does justfiy an update.

This is extremely minor, but enough that an update will make an audible improvement.

 

Have you noticed that these WONDEFUL results still have the vocal just a little too forward and a little too strident?   The fix is so trivial that it can be done directly in the command line on V4.0.5A (the current version), but I really do need to do a release and associated testing.

 

The equivalent correction on the command line is thus:

--pvdh=3k,3 --pvdh=3k,-6 --pvdh=3k,3 --bh=6k,-6 --pvdh=6k,6

 

Basically, on output it changes a 3dB 3kHz cut into a 6dB 3kHz cut and 3dB 3kHz boost.  (the shape of 6dB and 3dB EQ are different, so they cannot just be added together.)

For the pre-emphasis/de-emphasis, it adds an input pre-emphasis of 6dB of 6kHz cut, and a output de-emphasis of 6dB of 6kHz boost.

 

This difference is very noticeable, and is very minor in the scheme of things, but is also important for matching the original pre-FA encoded recording.

 

There IS a possibility that an alternate correction might be better, but this version appears to be the correct one.

GOT LOTS OF TESTING TO DO.   Sadly, I was about 2/3 of the way through a --xpppp ABBA decode, but gotta redo everything now!!!

 

John

 

[John Dyson]

Next decoder version is OUT OF THIS WORLD.   Beyond any previous expectation.

Basically, since I got some help with reliable directions, the decoder is technically correct now.  (modulo minor adjustments.)

This has given me time to work on secondary quality issues.

 

Was going to provide an example -- instead, it is probably best to wait for the release.

I am feeling REALLY GOOD right now :-).

 

The release will probably come about +12Hrs. 

 

John