'FeralA' decoder -- free-to-use

[BVanL]

Hello Mr. Dyson,

I am hoping you might be willing for me to contact you directly. 

 

I have started looking to replace legacy NR hardware from the late 60's, and the version from the late 80's, that will only become harder to maintain and support. I reproduce content with known, calibrated operating levels.

 

Kind regards.

BVanL

[John Dyson]

9 hours ago, BVanL said:

Hello Mr. Dyson,

I am hoping you might be willing for me to contact you directly. 

 

I have started looking to replace legacy NR hardware from the late 60's, and the version from the late 80's, that will only become harder to maintain and support. I reproduce content with known, calibrated operating levels.

 

Kind regards.

BVanL

I'll send you a private message...

 

[John Dyson]

You'all probably realize still working on the decoder.

Found out more about the phase descrambler.   Earlier versions of descrambler were missing some 'tricks' that help to further structure the signal.   There are some really, really devilish details in almost every step of reverse engineering the decoder.   Every time it seems like the program might be close to completion, then another layer of complexity is found.

 

It might be another few days before ready for the release, but GOOD progress is being made.   The development needs to be done at a very leisurely rate for now.

[John Dyson]

Release status:

About 2-3mos ago, finally found a way to measure the 'flatness' of the decoder, and that was good.   Now, in order to better 'decode' recordings, an different kind of dynamics expander was needed & added.  This is VERY different from a normal expander in the traditional sense, and it adds up to about 2.5dB to HF peaks.  This recovers the peaks that were originally in the recording.

 

Here is the problem, and the current challenge:   Those new peaks screw up the 'flatness' measure, making the HF seem to be approx 2dB stronger than it is.   It is amazing, if the HF has the same level (per my measurment) as the LF, now the HF becomes buried instead of balanced.   It really does add a dB or so to the peaks.

 

The challenge in the last week has been dealing with this new HF dynamics expander (same as the phase descrambler), finding correct settings and doing the 'right thing' to end up with a correct response balance.

The ONLY saving grace is that a 'correct' design will naturally provide this 'true, desired flat' response that isn't really flat.   What is this 'correct' configuration?   These issues have been the challenge over the last week or so.  On top of that, I have been trying to do a release tonight (for personal reasons), but unfortunately I have lost my HF hearing, so cannot double check the results.  The HF hearing will likely be recovered after I am available again on Friday morning (or hopefully Thursday night).   At that point, the final release tests will be practical, and MAYBE a release later on Friday.

 

 

[John Dyson]

Working hard to make the release available, but a long standing issue really does appear to be a bug.

 

With all of the details about the phase-descrambler coming into focus, it appears that alot of frequencies in the design are related to the 221.5Hz base frequency of the descrambler.   This 'base' frequency also includes some of the steps of 'bass' equalization.   These frequencies are being reworked RIGHT NOW.   Still trying for a release before the weekend.

 

The difference will be a more full bass, a little bit more simialr to the FA sound, but still MUCH MUCH more clean sounding.

 

 

[John Dyson]

What is delaying the release?

 

This delay is mostly related to the *NEW* phase descrambler, which also acts as an HF expander.   There are HF phase descrambler tenticles all over the decoder because there is a change in some of the base EQ frequencies.  This addition of the HF phase descrambler has some mechanisms to mitigate various kinds of of distortions,  all of the frequencies must be a multiple or submultiple of the base 221.5Hz.   Note that previously an important frequency sequence of 25,75,250, 500, 750 is really 221.5/9, 221.5/3, 221.5, 221.5*3.   Also, there is a criss/cross 2nd order EQ needed in the LF.   Included in these changes is that most of the EQ below 1kHz is 6dB 1st order instead of 3dB.   After making these LF changes, a flat LF comes very naturally with few real bumps.

 

At the end of this posting, the LF EQ list is shown.  Small bumps are meaningless, the important thing is the general trend.  However, the measurements below are the 'high res' version, so there is more error in the numbers, but show more detail.   The 'normal' measurement shows much more 'flatness'.

 

The HF descrambler is a complicated thing in its own right.  Also, my assumption about the standard frequency sequence was WRONG, but once the descrambler was reverse engineered, the new LF EQ numbers could be determined.  There are LOTS of interacting changes, but the new 'numbers' are pretty much figured out.

 

My only worry is that the BASS/HF ratio is very similar to the FA RAW copy, but I am not 100% sure that the HF average level is strong enough to match the bass.  Right now, my hearing is over loaded, so I cannot tell yet.

 

IF YOU ARE INTERESTED IN THIS THING AT ALL, PLEASE GIVE THIS UPCOMING UPDATE A QUICK LISTEN.  Sorry about missing the phase descrambler for >3-4yrs though -- finding out about it came from desperation about the HF needing more expansion.  I couldn't make the normal dynamics processor do the HF expansion and trying to compensate for missing HF expansion was ONE of the reasons for too much HF in the past.

 

The new update will NOT be available until all of these new changes work perfectly.  A new full demo-decode is being tried right now, and does sound good.  Gotta wait for a while, listen again, etc, etc.  If the decoder continues being okay, a release might happen tonight.  It make take a few more experimental iterations, but it is looking VERY GOOD!!!

 

High res LF EQ measurement.  Value is NOT important, but the changes are important.   The nominal gain value might be -3.7dB.   Even bounces of +-0.25dB are measurement error, but there are a few EQ errors that go beyond that.   For example, the 100Hz range still has a small bump, but is possible (not probable) a measurement error:

 

dB raw: -47.91 dB dec: -51.57 dB diff: -3.66
LEVELS 1300Hz to 1400Hz
dB raw: -47.66 dB dec: -51.37 dB diff: -3.71
LEVELS 1200Hz to 1300Hz
dB raw: -47.35 dB dec: -51.11 dB diff: -3.76
LEVELS 1100Hz to 1200Hz
dB raw: -46.89 dB dec: -50.69 dB diff: -3.8
LEVELS 1000Hz to 1100Hz
dB raw: -46.24 dB dec: -50.05 dB diff: -3.81
LEVELS 900Hz to 1000Hz
dB raw: -45.39 dB dec: -49.20 dB diff: -3.81
LEVELS 800Hz to 900Hz
dB raw: -44.44 dB dec: -48.23 dB diff: -3.79
LEVELS 700Hz to 800Hz
dB raw: -43.52 dB dec: -47.27 dB diff: -3.75
LEVELS 600Hz to 700Hz
dB raw: -42.68 dB dec: -46.38 dB diff: -3.7
LEVELS 500Hz to 600Hz
dB raw: -41.65 dB dec: -45.31 dB diff: -3.66
LEVELS 400Hz to 500Hz
dB raw: -39.98 dB dec: -43.65 dB diff: -3.67
LEVELS 300Hz to 400Hz
dB raw: -38.00 dB dec: -41.70 dB diff: -3.7
LEVELS 200Hz to 300Hz
dB raw: -35.83 dB dec: -39.45 dB diff: -3.62
LEVELS 150Hz to 200Hz
dB raw: -36.66 dB dec: -40.25 dB diff: -3.59
LEVELS 100Hz to 150Hz
dB raw: -35.89 dB dec: -39.84 dB diff: -3.95
LEVELS 50Hz to 100Hz
dB raw: -36.02 dB dec: -40.48 dB diff: -4.46
LEVELS 40Hz to 80Hz
dB raw: -38.20 dB dec: -42.53 dB diff: -4.33
LEVELS 20Hz to 50Hz
dB raw: -46.50 dB dec: -50.35 dB diff: -3.85
LEVELS 20Hz to 30Hz
dB raw: -61.17 dB dec: -64.79 dB diff: -3.62
LEVELS 10Hz to 30Hz
dB raw: -60.82 dB dec: -64.47 dB diff: -3.65

 

[John Dyson]

While listening to the outputs of the 'production line', the sound is good/clean using three very different sets of headphones, but impossible for me to tell much by speakers.

 

 THE RESULTS SOUND TOTALLY DIFFERENT FROM THE PREVIOUS VERSIONS.   With the new 221.5Hz frequency base instead of 50/75 Hz (250Hz) frequency base, the sound appears to line up much better and DOES sound more natural to me.  Also, the descrambler really helps, esp since the 221.5Hz frequency comes from that processing section.

 

An early comment about the response balance:   The natural LF/HF balance is different than previous.   There is naturally a bit more LF, but the LF/HF matches better, so the LF does not obscure the HF like it used to.   It is very possible that the decoding results might benefit from a slight LF rolloff at 45Hz (until now, I had always wondered why 45Hz worked so well -- 221.5/5!!!)   The HF is really good.

 

This bass balance is one reason why I am still holding the release.   The response needs to be carefully evaluated and some explanation why the result sounds so good, but is a bit bass heavy.   The bass response IS flat (modulo the slight 100Hz bump.)   Because of my worries, it appears that the 100Hz bump needs to be eradicated, and see if the sound is a bit less bass heavy.  IT IS VERY POSSIBLE THAT THE SOUND IS CORRECT, because it sounds normal on two sets of headphones, but is a little 'flat bass' heavy on a third.  Since the third headphones are by far the most 'flat', it is best to trust them, and indeed estimate the bass to be a bit 'interesting'.

 

Until the LF/MF/HF balance is less 'INTERESTING' and more 'VANILLA', or if there is convincing evidence that it is correct, the release will be held back for now.   Later on, if there is an increment of progress in either decision or decoder change, I'll make the decoder/demos available to the reviewers first.  There is a serious need for feedback on this one.

 

The decoder might actually be essentially flawless, and my brain cannot accept it.  Until there is stronger evidence that the decoder is or when (soon) it is correct, the release MUST be held.

 

 

[John Dyson]

When doing the last minute listening session to all of the demos (I actually listen to most of the recordings, not just the snippets), I found two bugs that are easily fixed, so will delay again.  The more active reviewers are already informed, and they have access to about 1/2 of the snippets.  I decided to give public access to the snippets also -- simply to show the progress being made.   DO NOT even bother downloading a copy of the decoder, because there is a better one being created today,speedup tomororw,  hopefully for the decodes completed in 2 days.

 

Here are the incomplete snippets for the V6.0.6G non-release.  I'd suspect that most would find them to be pretty good.   I know that a new pass will be better:

 

(Look into the sub-directory for the V6.0.6G snippets)

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

(The V6.0.6G decoder is also in the usual spot, but probably not worth bothering right now.)

WIth the recent questions about the DolbyA mode, I have decided to upload the V6.0.6G decoder, even though isn't ideal yet for FA mode.

 

If there is yet another day or two delay, it might be possible to improve the performance by fixin the recently noticed speed bottleneck.   There is a thread imbalance where the final EQ is a very complex beast, and is taking enough CPU to create a bottleneck.   The post decoding EQ has approx 500 1st order and 2nd order equalizers.   It is a very very daunting piece of software, and ended up being larger and more complex than I had originally guessed.   The speedup is likely very easy from a raw programming standpoint, because the equalizers were originally not laid out to be efficient -- because using only 10 equalizers is no big deal.  But, with 500-1000 equalizers, even the cache effects are important.   My 10 core machine with a huge cache close to the CPU has no troubles with the post decoding EQ, but my more normal i4770 4 core machine chokes badly on it, running very, very slowly.

 

Figure that there will still be a 3 days delay, but will include a first pass attempted bottleneck speedup along with the very slight EQ error fixes.  (very slight sibilance shift, mistake in the MF/HF join -- very minor, but messes up some transparency, and mistake in the demo decoding commands.)

 

[fas42]

Just had a listen to the Super Trouper snippet ... 👍

[bogi]

I went through V6.0.6G snippets and I am pleased. 🙂 I am able of more critical listening on later evening hours when everything around is more quiet, but what's clear for me, you solved the bass issues of end of 2021. For example the foot drum at the beginning of Tears Of Heaven was not heard in your demos few months ago but it is clearly here in the current snippets. The overall tonal balance and instrument timbres IMO now very well correspond to the original recordings. A bit of unsure feeling I had in the guitar presentation of Why Worry if the string attacks are not too stripped and therefore a bit edgy - but since nobody knows the production master, it may be correct. The other tracks I tried I liked without such an objection. 👍

[John Dyson]

4 hours ago, bogi said:

I went through V6.0.6G snippets and I am pleased. 🙂 I am able of more critical listening on later evening hours when everything around is more quiet, but what's clear for me, you solved the bass issues of end of 2021. For example the foot drum at the beginning of Tears Of Heaven was not heard in your demos few months ago but it is clearly here in the current snippets. The overall tonal balance and instrument timbres IMO now very well correspond to the original recordings. A bit of unsure feeling I had in the guitar presentation of Why Worry if the string attacks are not too stripped and therefore a bit edgy - but since nobody knows the production master, it may be correct. The other tracks I tried I liked without such an objection. 👍

Thanks,  I am very thankful for your specific, conditional criticism about 'Why Worry'.  I am going to seriously consider any change in results for that recording during the upcoming improvements.

 

We are getting CLOSER to perfection, and most of the time the decoder is doing pretty well.   I am a perfectionist, even though earlier releases might not suggest that.   I think that your evaluation is spot-on, and there is another apparent 'quantum' improvement coming in the next day or so.   There is still a slight challenge in choosing the correct way to get the perfect results on some of the Carpenters recordings, but the 'perfect' results are attainable.   (The reason for my mention of any specific group or recording is that since there are no specifications, a lot of tests across numerous genres and recording styles is needed.)   Carpenters, ABBA, many 'classical' recordings, each have a certain, special recording style.   There are others also.   This is an extreme back-and-forth, back-and-forth, over and over again to catch and detect all of the deviations from desired results.

 

New releases are very carefully derived from 'V6.0.6G' with no large scale deviation from it.   There is one specific upper midrange/lower HF item that will 'lock-in' the almost perfect sound.   It is a slight rebalancing of the 2.75kHz/3kHz region, and it is VERY, micro-slight change.   Once the best way that the magic setting is found, then the sound 'locks-in'.   There is also a slight improvement found in the phase-descrambler, and that correction only improves the sound on complex chorus.   Also associated with complex chorus, the DolbyA module still had a problem, which is now improved.   Greatly improves the chorus on the Olivia Newton John's recordings.

 

ALL changes are now very slight, and even though the next release is intended to make additional improvements, helpful criticism such as you have suggested is so very helpful!!!

There are no impediments to my full attention for the next few days, and a new release, after careful testing, will be coming in days.  Again, I regret this recent delay, but it is so important to  avoid botches like I have caused in the past.  (This requires carefully considering the variation in my hearing, and this does necessarily slow down some of the progress, forcing much more careful, very carefully considered change.)

 

Thanks again!!!

John

 

[John Dyson]

Very soon, starting the first 'full demos' of the next release.   Most often before a given release, there are at least 3 iterations catching minor errors, but not usually major redos.

 

The expected changes/similarities:

 

1)  More detailed/deeper modulation of the highs.   Basically, the same highs, but more dynamics without 'boosting' the HF EQ all that much.  These HF dynamics are actually a more complete application of the phase-descrambler, which would be thought of as an HF expander, but acts similar to a direct nonlinear expander instead of an attack/release type expander.  But, to reassure the reader, there is no distortion of sine waves, but instead the descrambling is done by non-resistive/non-lossy mechanisms.   Currently, the sound of chorus isn't quite as good as the 'G' release, and still finalizing the correction.

 

2)  Should be similar bass, even though the EQ needed to change caused by the major processing improvement for the highs..  The highs needed some 1st order change so to further correct the 2.75kHz/3kHz region, but 1st order EQ has skirts/leaks over wide bandwidths.  This leakage manifested down into the upper LF range, thereby required rework of the LF EQ.   The LF still has a favorable response curve.  The LF response curve will probably be further improved before release.  (The HF response curve is impossible to determine, even using the non-sine driven technique being used now.   There is an estimation that the HF has the correct curve based on design patterns and A/B comparisons.)

 

3)  Merging most of the pre-emphasis/de-emphasis and post decoding EQ into chains instead of separate EQ steps.   There is much less read after write activity, which can really screw with older CPUs.   Cache locality and sequential memory access is also much improved.  The should be a noticeable improvement on less sophisticated CPUs like the i3000/i4000 series.

 

Newer CPUs, including recent Ryzens and i10/i11/i12 should not have been severely impacted by read-after-write.  But, machines with normal sized (256k on intel) 2nd level cache should also see some improvement.  I believe that AMD CPUs tend to have much larger caches though, but improvement on i10 or newer and/or Ryzen2/3 should be less necessary.

 

One minor item of note...   The 'hollow' sound on the recording Carpenters/'Help' (track 5 on the 1970 album) is now much better reproduced.  (One of the 'tells' when debugging the decoder.)  Recordings like 'SuperTrouper' on ABBA *in the past* had a pronounced uneven sound, now mostly mitigated.   On almost all recordings, 'ticks' and cymbals are better reproduced, but little or no 'burning out one's hearing'.   (Earlier versions of the decoder had a very strong, notable intense sound -- mostly caused by an attempt to undo HF compression that unknowingly could only be corrected by the phase descrambler.)   I have heard NO overly  intense sound anymore.  *There are a few recordings with insane sibilance, both on the FA and decoded versions, only correctable by anti-sibilance.   Very, very few recordings now have the extreme sibilance problems.

 

 

 

 

 

 

[John Dyson]

It has been a rough day or so.

I might have unknowingly oversold the previous working/testing/debugging version of the decoder, evidenced by I have felt 'there is something wrong', including with V6.0.6G, which I DID like that version, just not perfect.   All positive feedback had also been conditional also -- I think that everyone was not totally happy yet.

 

Been spending the last day iteratively doing demo decodes/listening/reviewing/making changes/etc.

The basic design of the phase descrambler has been absolutely correct, but there was something wrong that the sound wouldn't 'LOCK-IN'.   The sound has been 'wobbly', and would drift all over the place when it comes to time domain and phase.

 

About 10 minutes ago, the formula for 'locking-in' has been found!!!   The detailed-details aren't all that important, but basically it was about trading some of the 1st order EQ for some 2nd order/Q=0.8409 (Chebyshev) type EQ.   Relatively simple change, but a lot of errors in the sound 'fell out', and became non-issues.   There are literally, not exaggerating, 1000's of details and at least 100's of architectural choices.  Without informed guessing (and reading dead people's minds -- just kidding), this would have been impossible to get as far as it is now working.

 

Given the fact that the last day or so has been spent working on a rather important 'misbehavior', the next release might be delayed another day or two.  LOTS of testing will be needed for the corrected phase descrambler, but it really does sound good.   "Dreamer' on Supertramp/Crime of the Century is amazingly better sounding, with the somewhat garbled vocal being centered into the stereo image.  (Basically, it appears to sound 'normal'.)   No previous decoder, include the misguidedly 'liked' single layer version couldn't correct that recording.

 

I am hoping for release tomorrow night +30Hrs or so, but if everything goes perfectly, and I am so buzzed about this, might be able to release sooner.

 

 

[John Dyson]

Sometimes the final, simple things take the longest amount of time.

 Currently on the 4th or 5th iteration for the correct MF/HF join setting.

 

The phase descrambler with it's up to 2.6dB change of HF level really confuses making the correct choice for the HF/MF join.

Starting the 6th iteration for the demos/tests.   Frustratingly, the errors sometimes pop-up in the last 1/3 of the recordings.  This really means that ALL of the demos need to be run though the decoder, and listen to a large portion of all of the recordings.   Also, part of the 'listening' is doing comparisions between the FA original and previous demos (e.g. V6.0.6G) as a basis.

 

A good review takes at least an hour or two, after the decodes of the demos are complete.

 

Just to let you know, there is still progress being made, however this last bit is very slow.

 

 

 

 

[John Dyson]

Good news...   There are two prospective versions that are (or soon will be) in the hands of the active reviewers/testers/commenters/etc.  Anyone interested in helping with the A/B choice, just let me know, and I'll make the correct comparison versions available to whomever interested.    This message is offered as 'status' info, and not 'release' info.

 

No matter what, the expansion on either version is notably complete/correct than any previous decoder version. After this choice between the prospective V6.0.6P or V6.0.6Q versions, then one or the other might be the release.   Until it is announced, the release is not ready.   In fact, it is possible that neither P nor Q will be the 'release' version, but instead a non-existent 'R' version might need to include other corrections.

 

Moving forward!!!

 

 

 

[John Dyson]

Release delayed to +6Hrs.

It is working fine, but still working on choosing the correct 'join' between MF and HF...  (Tricky.)

 

 

[John Dyson]

Sorry, still having troubles.   My hearing is making it impossible to test/verify the output.

 

Will keep you up-to-date.   My intention was to send this thing out about 3-4Hrs ago for UK/EU people, but frustration is the name of the game right now.

 

In desperation, a release was just started again, with ALL of the phase-descrambler EQ steps enabled.   It sounds correct, but only based on what I hear now.   After 1/2 way through the demos, I'll review again.   In 2-3Hrs, the results could sound very bad after it sounds perfect right now.    

 

It is VERY possible that using all of the descrambler building-blocks is correct, just that I cannot audibly verify it.   I'll make the assumption it is correct if the results are plausible this time.   I'll do the release without being able to know if it really sounds okay, just that it is probably correct.

 

This is the 5-6th attempt based on this version of the program.

Can you imagine how frustrating this is for me?   The EE/DSP stuff is trivial compared to the problems with my broken hearing!!!

 

 

 

 

 

[John Dyson]

ADDED AFTER A QUICK REVIEW (AGAIN):

There is an error in V6.0.6V that (sorry again), I couldn't hear.   The graininess that was so worried about DOES exist...

The bugfix is simple, and will be updated within hours.

 

I DEEPLY REGRET THIS, but at least the decoder is finishing up, and is capable of being beautiful.

 

(The rest of the message below is pretty much accurate.  The announcement of V6.0.6W will be short and sweet.)

 

 

 

V6.0.6V is available.

This release has the phase descrambler fully (hopefully) enabled, speed-up for slower/older machines using simple decoding modes, numerous other bug fixes.

 

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

 

This release has been delayed because it is almost impossible to achieve a reliable review of the results by the author (me.)  It is likely that this is a functional release, but still not 100% sure about descrambling the 3kHz band.   If this works well, it should work *VERY* well.   The delay & tricky nature of the descrambler  is the only reason why the reviewers were bypassed on this specific version.   The descrambler errors can be very subtle, and it is very important to know what to listen for -- it is too much to ask the reviewers/testers to have spent a lot of time (many hours) of training to learn the 'descrambler error' sound.   Great care has been taken, but still there might be a problem with the join between the MF/HF and the effects of the phase descrambler.  Now, there is almost 100% probability that finally the MF/HF join is correct.  On the other hand, the descrambler might be perfect, or might need a change in ONE major setting along with a few minor collateral changes in the settings.

 

A simple heads up about the possibility that HF might have 'crinkle'  or 'grain' in the sound.   I cannot hear well for enough time to properly test for it.   IF THERE IS A KIND OF WEIRD GRAINY SOUND IN THE HIGHS, let me know!!!   That means that the 3kHz band must be enabled in the phase descrambler, but it is tricky to know whether or not to enable 3kHz.   Tell me immediately if you hear the 'crinkle' or 'grainy' sound in the highs, and I'll immediately correct the problem.  Enabling the 3kHz band without it being necessary can cause very serious problems, but if needed -- then will correct all kinds of problems in the sound.    This has been VERY FRUSTRATING for the last 2-3days, where if the decision could have been made, the release would have been made 1-2days ago.   Changing the 'descrambler bands' utilizes the building-blocks nature of the decoder.   * when the supporting frequencies for the existent very high frequencies are missing, the resulting sound can sound 'grainy', and this is what I am worried about.

 

What is the 'phase descrambler'?   It can be looked at as a high frequency super-fast dynamics expander or lining up the phase of the HF signals so that they are more coherent.   The scrambling effect hides peaks in the signal, and the descrambler restores the peaks without nonlinear distortions.   (it really is a phase scrambling/descrambling scheme.)

 

Numerous other quality improvements have been made, including the correction of some expansion effects that result from slight DA decoding timing errors.   The very slightest error in DA decoding can make very noticeable botches in the FA decoding.

 

Since my hearing is good for only 4-5minutes at a time now (then must lay down for 30min to 1Hr and 'recharge' my hearing again), it has been almost impossible to verify the correct descrambler settings and the correct configuration of the MF/HF join.    The good news is that the 'join' has an almost 100% probability of being correct/perfect, but the descrambler is still a bit of a problem.  There are many interactions, and choosing the right audio bands to be controlled by the descrambler has been a daunting process.   After a lot of testing, 3kHz would have been nice, but could cause all kinds of problems.   It has been challenging to determining whether the 3kHz region has been been 'phase scrambled.'   6kHz appears to be phase scrambled X 2 (two passes of scrambling), and 9kHz appears to be scrambled X 1.   The rest of the bands, 12kHz, 18kHz, 24kHz and 30kHz do not appear to need to be descrambled, even though initial results when enabling these bands superficially sound correct, but then the problems from over-expansion start appearing.   If the 3kHz band does need to be enabled into the descrambler, then the use of 12kHz through 30kHz needs to be revisited.

 

If 3kHz is supposed to be included, then in this (V6.0.6V) release,  the higher bands will have a 'crinkle' in the sound.   If anyone hears the 'crinkle', then A new release needs be produced IMMEDIATELY.   It is just that the tells for the 3kHz band are confusing with poor hearing, and simply knowing that it is correct to add the 3kHz band into the descrambling will make it MUCH MUCH easier to add the band and change the other settings to accommodate it.

 

The ONLY reason why the release has been delayed, and I have asked about the 'grainy' sound is that the last 1-2 days have been spent on the phase descrambler, and I feel that it is time to make the decoder available, then IF there is feedback about graininess, then the decoder can be fixed in minutes.  (The demos will take a few hours though.)

 

THANKS, and I really hope that the worrisome 3kHz/phase descrambler problem is non-existent!!!

Otherwise, once this descrambler problem is determined existent or not, the decoder is pretty much complete (except for minor changes, if needed.)

 

 

 

[John Dyson]

Further status on V6.0.6V/V6.0.6W

 

A reviewer caught the demo uploads as they were happening.   After doing the long announcement message, I got some feedback about a somewhat heavy, dour sound in the highs.   IMO, that is a way of describing the sound character that had been worrisome.   My regrets are VERY strong about this release and the troubles that I have caused.  Trust me when I say, I admit that the release has been flawed, but the decoding results can be unbelievably intriguing, if not beautiful.   For me, for enjoyment sake, it is best to casually listen to the results, otherwise the details overwhelm the actual beauty of the recording...   It is SOOO important to ignore the details, but listen to the music!!!  Unfortunately, the details need to be understood and problems resolved, and I am the only one that can do it.

 

After accepting some feedback on the problem, also requiring some translation from 'Audiophile speak' to 'Engineering speak' (two different languages :-)), a prospective  change has been made.   It will require approx 1-2Hrs to be able to determine if the sound is 'better' or 'worse'.   During this 1-2Hrs of waiting, the demos decoding will be running.  It is very likely that the sound is very noticeably (if not profoundly) better.

 

The demos are being decoded at a more practical level of quality '--xp=max' without '--dp'.   A few select demos will be done at '--xpp=max --dp'.  For normal, recent 4 core machines, '--xp' is probably good enough, but if doing a super-high quality commercial release (forgetting about licensing for now), '--xpp=max --dp' gives the very highest quality performance.   Any mode like '--xppp' or '--xpppp' is very likely to cause strange, weird damage that is almost indescribable.   The '--xpp' mode doesn't reach down far enough to damage important sidebands created by instruments or vocals.  '--xpppp=max' is liable to suppress the instruments in a recording, but it will end up being VERY clean sounding -- too clean :-).

 

 

[John Dyson]

Trying for a release attempt again later on today.   It is very close to 'perfection', but not quite 'it'.   I better understand the previous feedback given, and isn't quite what I had thought that it was.  Running tests right now, and should have the best corrections soon -- hopefully no later than early tomorrow (hopefully +12-+14Hrs), but this is very complex stuff!!!

 

Frustratingly, because of the lack of spec, it is difficult to distinguish between a 'correction' and a simple 'modification'.  Being as careful as possible.

The good news about the internals -- the 3kHz range could be added now, which does fix some problems that popped up a day or so ago.

 

With the new group of improvements, some of the code has 'dropped out', and that is a good thing for sure.   One must remember that the design of the FA encoder was based on HW, therefore the correct design is likely maximally simple.   Any time that complexity can be removed in a consistent/valid way, then it usually means that the result is more accurate.

 

 

[John Dyson]

The problem with the HF phase descrambler/dynamic expander has been vexing, very frustrating.

 

It is so very helpful and appreciated when constructive, but critical feedback is provided.  The well intentioned critical feedback spurs on improving the decoder.   Sometimes, finding new bugs on released code does elicit a modicum of panic, but soon goes away.

 

After these several weeks figuring out how to use the descrambler, there is now a good understanding about using it.

 

1)  Make sure that the decoder without the descrambler is nominally flat.

2)  Adjust the settings in the descrambler so that it appears to somewhat  subjectively match the FA version, but with more convincing dynamics. (and other corrected features.)

 

It is folly to try to measure the static spectral energy response after the descrambler is installed.  Regrettably, it appears that the basis of adjustment is currently limited to subjective evaluation.  The objective methods for measuring the response are normally poor quality, but there is even less of an accurate objective measurement after the descrambler is installed.   Since  using the descrambler is much better defined than before,  a command line method for evaluating the enable/disabled sound is now more useful.

 

Once finding that it is important to start with the decoder being nominally flat, some new modifications to the descrambler were very helpful.  The descrambler needed to be changed to be adjustable for each subband instead of a single across the board setting.

 

Currently, in the experimental code, there is a separate setting for each of the normal processing bands of 3k, 6k, 9k, 12k, 18k, 24k, 30k.   Dynamically expanding at 3kHz is still in question, but the other bands appear to be necessary to 'flatten' the sound.  The ONJ "Country Roads" recording is very important for the tells and ability to clearly adjust the settings so that the chorus sounds balanced and relatively normal.  (Before descrambling, the 'Take me home, Country Roads" sounds very mangled.)

 

Once adjusted,  using new descrambler produces a more smooth sound.   The final adjustment isn't determined yet, but the 'knobs' are  usable understandable.   As of now, the values for the 'knobs' in dBs have been chosen from the standard sequence based upon multiples/submultiples of of 3db (e.g. 1.5, 0.75, 0.375, etc).   There hasn't been a need to use >=3dB, but 1.5dB appears to be the nominal value for 'sounding right' for the 6kHz and 9kHz bands.   The 3kHz band is problematical, and might force the use of negative dB values.  (These dB values are the dynamic store/forward values for the descrambling actions at each frequency band.)  It is unlikely that the 3kHz band will need expansion.

 

Since the descrambler settings are understood and once accurately determined, there will be a chance for a much better release.   As mentioned in the previous paragraph, there is a notable improvement, but also must pace my subjective review because of limitations in hearing ability.

 

Very often, when comparing new releases vs earlier versions, it elicits a LOT of embarrassment about earlier versions.   In a way, that is a good thing that progress is being made, but sadly, the decoder never seems to be perfect.   We'll have to solve that problem of imperfection, and we are getting closer and closer.

 

 

 

 

 

[John Dyson]

The last day, been trying to find the optimum settings for the HF descrambler, but not confident in any of the choices.   This setting is NOTHING like a matched spectral response on typical recordings, but instead is a kind of 'enhancement' over and above the FA signal.   Of course, the goal is NOT to enhance the original clean signal before FA encoding.  The goal is ONLY to recover the original pre-FA signal.

 

The 'rub' is, where is the point where FA scrambling is undone, but nothing bad is done to the pre-FA signal?   As of these last 24Hrs, I just don't know.  The settings are 'clean/orthogonal', and appear to have the correct type of settings.  There are  two places/settings, very similar, where the sound becomes smooth and less 'grainy'.   Choosing the singular correct setting has been challenging.

 

If this continues to be a confused goal by tomorrow, instead of the descrambler being enabled all of the time, it will become an option in the tomorrows/tuesdaysrelease.   With that option, the best setting will be default.  It doesn't seem to be a good idea to allow choice in the descrambler settings, because it complicates usage.

 

The tuesday release is the intended release on Friday, but with the descrambler matter resolved one way or another.   The defects in the sound character on the Friday release were 100% caused by the descrambler and mistakes in the settings.

 

There is still a chance of resolving the descrambler setting matter before tomorrow, and importantly, the descrambler WILL be available in the decoder.   I just don't know the ideal setting, and the goal is 'first, do no harm'.   The descrambler is very capable of doing harm, so must choose the best settings.

 

 

[John Dyson]

Working hard on the HF descrambler, but the results are good...   The problem, which setting is most correct?

 

The results are:  

* 1.5dB higher peak-average ratio (in current mode settings.)

* Allmost no change in the average signal level (in hundredths of a dB).

* When enabled, the descrambler output practically always sounds at least slightly more clean/detailed than without.

 

With the corrected & better integrated descrambler, it is an 'independent' module of sorts.   Changing the mode to enable/disable the descrambler places no requirements on the rest of the program.  (No gain or EQ or mode changes needed beyond enabling the descrambler.)

 

Other than finding the singular, absolutely correct setting, the most worrisome problem is that the bass sometimes *seems* to be a little more weak when the descrambler is enabled.   However, when measuring the bass energy (both peak and RMS), the results are the same (hundredths of a dB) between descrambled or not.   These results are good, because the bass energy is not expected to change (much.)  Even the most aggressive mode, the descrambler base EQ frequency is 3kHz.   It does appear, however, that 6kHz as a base frequency is less disruptive, and sounds more natural.    Due to several reasons, the 3kHz base freq appears to be wrong, so the sequence will likely start at 6kHz, even though it would be a good idea to try 4.5kHz just in case.  (Note:  the average energy is not expect to change much, whether bass, MF or HF... )

 

If the descrambler uses the incorrect base frequencies, the quality quickly diminishes.

The challenges are still:  being very sure about the 6kHz base frequency, being sure that the 'depth' of the descrambling should be 1.5dB.

 

Interestingly, the depth of the descrambling per base frequency is almost the same as the increased peak-to-average ratio.   This bodes well for the correctness.  Sometimes when the peak-to-average is increased, there is not necessarily a pronounced change in the character of the sound.   On some recordings, the improvement is concentrated in instruments like cymbals'.   While optimizing the settings, using 'tells' like the sound of cymbals has been helpful.

 

 

 

 

 

[John Dyson]

Potentially interesting sound of the decoding results with the current HF phase descrambler enabled.

It is hard to describe the sound, and some of the initially apparent characteristics are NOT what is happening in the recording.

This is NOT a final version -- hopefully, there wont' need to be a 'descrambler disable', and it is getting close to not needing the 'disable' command feature'.

 

The best way that I can describe the more complete descrambler -- more open, happy sound.   Detail is astounding, if one listens carefully.

The general measured stats of the descrambled recording is very similar or better dynamics than the original decoded versions.

Even though one might guess that the HF or LF balance is different -- oddly, the energy level appears the same HF vs LF.   I am still trying to figure out if there is a slight 'tilt' towards the HF, but both the descrambling algorithm and the measurements are saying there isn't one added by the descrambler.

 

The demo version is 'V6.0.6XB'.   There haven't been private & requested versions uploaded, simply because they are large and the results are necessarily transitory.

There are only snippet demos available:

 

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

 

As soon as the descrambler settings are fully understood and that the settings UNDO the original scrambling, then there will be a release.

 

[fas42]

Just to give some immediate feedback, John ... again, the Super Trouper snippet. The voices intro lacks sparkle and air; and the following instrumental run sounds, 'artificial' - the following seems OK ... this was comparing it to the RAW sample, which I downloaded some time ago.

 

If you want me to elaborate more, let me know ...