Jump to content
IGNORED

'FeralA' decoder -- free-to-use


Recommended Posts

5 hours ago, StephenJK said:

John,

 

For purposes of comparison, I've dropped Yes - Close to the Edge and Genesis - The Lamb Dies Down on Broadway to this DropBox link:

 

https://www.dropbox.com/sh/77crt0gvyofdqaj/AAAb4xUBq6fo50_sNviUnHhGa?dl=0

 

Both are from the 1973 and 1974 LPs that I recorded.  That may give you an analog reference.  Note that Close to the Edge is from Yessongs, and may not have the same progression as the studio track.  

 

I seem to recall we did this once before with a Supertramp track.  All my LPs were recorded on a much better than average turntable and MC cartridge, but of course can only be as good as the source material.  

 

I didn't check 'Close' yet, but your Lamb is DEFINITELY a 'pure', non-FA recording.  It passes my own review and the review of my program as not being FA.

 

THANK YOU SO VERY MUCH.   I have so very few good examples of pure recordings that also are available in FA form...

 

John

 

ADD-ON:  after listening longer and more carefully (tinnitus is driving me crazy right now), a the start of the vocal, there is some dynamic range compression -- during at about 30 to 35 seconds, but that is a GOOD thing in a way.  it shows that the record was *mastered* and not just FA compressed.   It is so nice to hear a normally processed recording, just like they used to be!!!

 

 

Link to post
Share on other sites

I know that I said that I'd put the decoder aside for a little while, but someone nudged me and reaffirmed a decoder bass problem that I was worrying about.   I wasn't sure, and simply accepting the 'tubby' bass.   This behavior was happening because of my 'rule' against 2nd order EQ.   With 2nd order EQ, the problem is almost trivial to fix, but I had been trying to make the bass work with a very minimum of 2nd order EQ.   Just so happens that another level of 2nd order EQ (up to three now) is needed to bring the bass up to where it should be, and NOT be tubby.

 

The upcoming release in a few days will include the improved bass (again), but will have a few more speedups.   Now, on faster machines, you can do a decode in the high quality mode '--xp=max' as quickly as '--fz=max' in older versions.   The big difference is that the --xp kind of anti-distortion is basically a full magnitude better than '--fx and --fz'.   The '-xp' series of anti-distortion modes include: '--xp', '--xpp', and '--xppp'.   The best normal mode is '--xpp', because '--xppp' is simply insane, mostly for testing, even on decoding DolbyA materials.

 

The bass SHALL be fixed.  Thank you, anonymous person for the encouragment in the right direction!!!

 

John

Link to post
Share on other sites
3 hours ago, John Dyson said:

I didn't check 'Close' yet, but your Lamb is DEFINITELY a 'pure', non-FA recording.  It passes my own review and the review of my program as not being FA.

 

THANK YOU SO VERY MUCH.   I have so very few good examples of pure recordings that also are available in FA form...

 

John

 

ADD-ON:  after listening longer and more carefully (tinnitus is driving me crazy right now), a the start of the vocal, there is some dynamic range compression -- during at about 30 to 35 seconds, but that is a GOOD thing in a way.  it shows that the record was *mastered* and not just FA compressed.   It is so nice to hear a normally processed recording, just like they used to be!!!

 

 

 

John, I don't how to be more clear on this, and I'm trying to help:

 

The opening of Close to the Edge (3.4.0 --fa=7) DOES NOT SOUND LIKE RUNNING WATER

 

I hate to shout but bluntly I've heard this played many times live, and the opening after FA decode is demolished. I suspect the harmonic structure is being severely damaged. If you insist that this is in fact FeralA then you need to fix it... it you can't hear what I'm telling you (and read the Wikipedia description, I'm not making this up) then your hearing is not sufficient to judge. The chimes/tinkles on the opening of Lamb Lies down on Broadway has the same problem, again can you hear it?

Custom room treatments for headphone users.

Link to post
Share on other sites
8 hours ago, StephenJK said:

John,

 

For purposes of comparison, I've dropped Yes - Close to the Edge and Genesis - The Lamb Dies Down on Broadway to this DropBox link:

 

https://www.dropbox.com/sh/77crt0gvyofdqaj/AAAb4xUBq6fo50_sNviUnHhGa?dl=0

 

Both are from the 1973 and 1974 LPs that I recorded.  That may give you an analog reference.  Note that Close to the Edge is from Yessongs, and may not have the same progression as the studio track.  

 

I seem to recall we did this once before with a Supertramp track.  All my LPs were recorded on a much better than average turntable and MC cartridge, but of course can only be as good as the source material.  

 

Yessongs is a live album so different than Close To The Edge album. The intro is significantly shortened and not a great example of the running water.

 

The versions I am listening to include the original CD, the Steve Wilson remixes (outstanding) and the SACD

Custom room treatments for headphone users.

Link to post
Share on other sites
13 hours ago, jabbr said:

 

John, I don't how to be more clear on this, and I'm trying to help:

 

The opening of Close to the Edge (3.4.0 --fa=7) DOES NOT SOUND LIKE RUNNING WATER

 

I hate to shout but bluntly I've heard this played many times live, and the opening after FA decode is demolished. I suspect the harmonic structure is being severely damaged. If you insist that this is in fact FeralA then you need to fix it... it you can't hear what I'm telling you (and read the Wikipedia description, I'm not making this up) then your hearing is not sufficient to judge. The chimes/tinkles on the opening of Lamb Lies down on Broadway has the same problem, again can you hear it?

Listen to the natural, LP version of 'Lamb'.  Listen to the decoded version with '--fa=6' or '--fa=7'.  Evaluate the CD/Download version also, and do the comparison.  *The actual goal is NOT what we like*, the goal is more like the 'LP' version/master tape (sans the 'personality' of LP media.)   Choose which one you like for listening.  I have been working on the EQ for a year or so, but you are talking about the 'dynamics', but that is all driven by DolbyA behavior and how the units are networked together.   I'll probably re-visit the dynamics for the exceptional cases AFTER the EQ matter is closer to resolution.

 

Currently, there is no solution for your legitimate criticiwsm, and addressing it just this minute right now would defocus -- the 'mental state' and memory requirements for this project are IMMENSE.  Focus is important, or otherwise I won't remember anything.  (When changing from portion of the decoder to another, there is a 'relearning' exercise.   Changing states too much makes it easy for my error prone nature to create more bugs.)  If "Lamb" is not FA, then it would probably be best to listen in the state as distributed.

 

The decoder dynamics are good enough to say that it is completed in that area, but might need special modes in the future.  Just pondering the general method to create the correct dynamics is a challenge.  Right now, there is a matter of the EQ, and that is coming along nicely (but too slowly, like everything else on the project.)   If the dynamics behavior are not desirable on certain recordings, then at this point, I *must* defer work on the dynamics for that certain niche of recordings *right now.*

 

IMPORTANT PARAGRAPH:

A lot of people have often been upset when corresponding with me, often because it *seems* like I dont' listen.   That isn't ususally true, but instead I have a LOT on my plate.  I remember almost all constructive criticisms and work on them ASAP.   If there is a criticism that I just don't know how to deal with immediately, then I must temporarily move it aside while working on the behavior for the majority of recordings.   I have been focused on the EQ for the last year or so -- it requires a LOT of energy to keep focus and lots of patience, not just for me, but those who are helping the project. *There are some cases where I haven't understood the criticism, and that has been caused by my bad hearing and/or my headphones altering my perception*   My personal and equipment flaws have been a major impediment in understanding some criticism for quite a long time.  I am trying to correct that RIGHT NOW.

 

Some of the details might be less obvious in the LP version when compared to the CD/Download FA version, but I dare say that the LP version is probably the closest to the master tape than anything we have (again, ignoring the eccentricities of the LP media/playback.)   The LP version was/is a 'godsend' towards the goals of the project.   Imagine the frustrationg that I have numerous master tapes, but NONE where there is a direct FA version available to compare with. 

 

**** The decoder is NOT about focusing  what anyone specifically likes, it is about returning a recording back to its original state.   However  sometimes it is difficult to figure out the original state.   Sometimes the original state is more or less desirable for listening.  I suggest that there IS a secondary effect that in a lot of cases, the properly working decoder will give a desirable outcome.  CERTAINLY, IT IS NICE WHEN A CORRECT RESULT IS ALSO THE BEST SOUNDING.   More than likely, the correct result will indeed be a superior version for listening, but not always.

 

For listening, using the decoder opens up options, it sometimes solves problems, sometimes not.   The LP (thank goodness) gave the project a REAL baseline of correctness.  (A few other examples of non-FA baselines have indeed been made available to me, but NONE was as clear/listenable as the LP as supplied by @StephenJK.)   I cannot judge the result based on 'sounds best'.    This general criteria that I cannot primarily be 'what sounds best'.

 

Trying to make 'sounding good' an EQUAL or even slightly less priority as 'correctness'  makes it really difficult to develop the decoder.  I *must* base the critiera of success on 'sounds correct',  'follow the rules', but  'sounds best' might sometimes be of lesser status.   Right now, for a *few* recordings,  'following the rules' doesn't give the results that some people hope for -- but the rules do get modified from time to time.

 

The upcoming bass correction will be an example that results from an 'exception to the rules'.

 

Next post will give a technical update!!!

Please be patient as there has recently been major progress once I found some basic reasons why there has recently been problems understanding some criticism!!!

 

John

 

 

Link to post
Share on other sites

I am so pleased to report an update on a few improvements:

 

Three useful matters:

 

BASS:

The bass problem has been giving me fits for a long time.   The currently release comes close, but the bass has a sometimes terrible, and definitely incorrect tubby sound.  This sound 'muddies' up the result with a bass-veil over the clarity of the recording.

 

This bass problem has been troublesome.   To correct the bass, there had to be another exception to the 'rules' about avoiding 2nd order EQ.   By using the 2nd order EQ, it appears to be the effect as originally designed into the system.   Eventually, by the time of the release in a few days, that 'muddy' sound should be remedied.   The 'muddy' sound is 'definitely' remedied in the current test versions, but there are other interacting aspects of the bass, some more work might be needed.

 

* I plan to make a test-only version available some time today, but I am offering it as a 'status report' and maybe feedback by the most energetic users, but not enough testing for a real release (e.g. the AVX512 orSSE3 versions might not run.)  At least, the demos should likely be reliable.

 

More complete decoding:

Some recordings appear to have a little bit of lingering compression.  As usual, I had mistakenly believed that the default '--fa' (which is actually '--fa=+7') would be the standard compression used on ALMOST all recordings.  That seems to be less true than I thought.   I found that the ABBA catalog and Brubeck's 'Take Five' are encoded as '--fa=+8'.  When decoding more properly, the sound definitely improves.   This kind of thing *might* be a problem on other recordings.   Will give status on this matter at release time.

 

 

Faster operation at high quality levels:

The highly desirable --fz, --xp and --xpp modes will be even faster than the original planned speedup.  Please don't assume that the higher quality modes will be fast, because they will definitely NOT be fast, but a noticeable speedup of almost 2X or so will be available in --xp mode.   Also,on AVX512 machines, all higher quality modes (including the lowly --fx) will be calculated in higher precision, even if the 'double precision' switch isn't enabled.  (includes many of the i12 or perhaps later i11 machines.)   The increase in precision is pretty low cost on such machines, but on SSE3A, AVX2 machines, the precision will be the same as before unless using the '--dp=yes' option.   Using the '--dp=yes' option will run about 2/3s as fast as the default, single precision mode.   This slowdown is also true of using the explicit '--dp=yes' mode on AVX512 machines.   It is just that the 'single precision' mode on AVX512 machine slips-in some 'double precision' accuracy where it helps most.

 

The last few days have definitely been interesting.

Since there have been so many things going on, it might be a few days before release time again.

 

 

 

 

 

 

Link to post
Share on other sites
55 minutes ago, John Dyson said:

Listen to the natural, LP version of 'Lamb'.  Listen to the decoded version with '--fa=6' or '--fa=7'.  Evaluate the CD/Download version also, and do the comparison.  *The actual goal is NOT what we like*, the goal is more like the 'LP' version/master tape (sans the 'personality' of LP media.)   Choose which one you like for listening.  I have been working on the EQ for a year or so, but you are talking about the 'dynamics', but that is all driven by DolbyA behavior and how the units are networked together.   I'll probably re-visit the dynamics for the exceptional cases AFTER the EQ matter is closer to resolution.

I'm really trying to help you here. I've hear these concerts live and know what the sound is intended to be. My hearing is quite good, it doesn't vary.

 

The "original state" of the intro to "Close to the Edge" includes a recording of "running water". The sound is unquestionably intended to be running water. This isn't controversial. You can make your explanation of compression and the math as complicated as you wish, and then come up with whatever tweaks and alternate undocumented settings that you wish, but at the end of the day it needs to sound like running water. 

 

Lamb: try turning up the volume and measuring at how many seconds into the intro you can first hear the tinkles/bells ... at how many seconds can you hear? Turn the volume up as loud as you like. FA decoding removes the sound. That sound is intended by the artist to be on the track. This should be basic -- it is not your right to say that this sound isn't intended to be on the track.

 

The problem with using the LP recorded version is that the background hiss is substantial and enough to obscure details that are low level. 

On my SACD/DSD version I can hear the tinkles start at 10 seconds, same with the CD version. After FA decoding they don't come in until 20 seconds and the timbre is "muffled" and chunky. The LP version comes in at 25 seconds or so with normal timbre.

 

If you can't hear these specific things, I don't know what to tell you. Why would I trust your hearing to tell me what is or isn't FA?

Custom room treatments for headphone users.

Link to post
Share on other sites

Have an announcement -- wow!!!  (not wow as in perfection, but wow as in work to make the decoder correct.)

 

The release is much closer, and was actually planning on some 'status report' demos yesterday, but used the wrong version of the decoder for the builds. In the build scripts, I forgot to change the decoder back to the experimental version.   Hopefully, there will be satus report examples today, and a release sometime in the next day or two.  (I need to do the demos for myself anyway, since I review the outputs as carefully as I have time.)

 

When using some reference materials (not necessarily publically made available), I found some very specific flaws in the decoder.  One flaw doesn't make me happy, because the correction will not likely be pleasing to most current listeners of recordings.

 

1)  I think FINALLY came up with the correct BASS EQ.   I kept 'adjusting' the EQ mindlessly carried forward from previous practice.   The new change is basically allowed by a relatively new addition to the code.   There is a 'regular' form of EQ used on the LF, and previously had not considered directly joining that EQ to the normal micro level EQ needed at 75 and 50Hz.   To make a long story short, the LF EQ is much simpler, and follows the notion that the old HW design must have been as simple as reasonably possible.   I'd guess about 3-4 equalizers have been removed.   With the change to the EQ, the general sense of bass is the same as before, but without the perception of 'tubbiness' and with less burial of the higher frequencies.

'Tubbiness' is basically created by too much LF (in the 1-2dB range) at between 30Hz and 70-80Hz.   To emulate the needed low bass, a simple 1st order design is too non-frequency-selective to avoid the 'tubbiness'.   The new version of EQ, done with carefully chosen 2nd order EQ also, mitigates most of the tubbiness while still passing the 'tubby' frequencies when needed.   This EQ is REALLY much simpler, less guessing.

 

2)  Some of the reference materials told me that dynamic sounding highs were made 'too dynamic' by the decoder, sometimes replacing 'dyamic singing' with a kind of 'blast/distortion' in the vocals.  That is now mostly remedied by choosing yet another pre-emph/de-emph pattern.  The basic freq response is same as before, but the vocal dynamics are less overly strong.

 

BAD NEWS:

Reference materials are telling me that the highs have been decoded at equivalent to a 1st order HPF at 9kHz/6dB (basically about 2-3dB too much at 9kHz, with a rising error to about 20kHz.)   When adding a 9kHz,-6dB 1st order LPF on the default output of the decoder, the sound becomes more similar to an LP.   Without adding the EQ, then the sound is more similar to FA.

 

ANSWER TO BAD NEWS:

I am not an idealogue and don't believe in perfect or nothing, but perfect should be attempted and made available if possible.

Now, there will be two EQ modes, where the LP emulation mode is enabled by --hfrl=lp.  The --hfrl switch can also be used, in the same command, for the already existent 'careful' HF rolloff scheme also.   I'd suspect that most people used to FA recordings will stick with the default, but to be more accurate to normal/ancient LPs, then the --hfrl=lp switch will enable that capability.

 

One thing for sure, the decoder NOW is capable of coming much closer to the sound of old LPS, either with or without the '--hfrl' switch usage.

Especially to those helping with example recordings:  THANK YOU VERY MUCH!!!

 

John

 

Link to post
Share on other sites

Yesterday, there was a pre-mature demo release attempt.   The HF response was slightly rising in the released version, and it was because of a mistake in the EQ (it wasn't one of those typeo mistakes, it was a real mistake this time.)   The mistake was actually kind of severe -- but sometimes unnoticeable.  When the mistake was noticeable, it could be very very ugly.   Basically, a rolloff of -3dB at 12kHz was missing.

 

The correct demos/evaluation recordings are now V3.4.3A.   My internal test version is at V3.4.3C, just making a change to support more user specified 1st order post decoding EQ.

 

There isn't a V3.4.3A decoder uploaded because the release is coming up very soon.  (maybe even hours.)  There are additional helpful bugfixes added since V3.4.3A.

The V3.4.3A were intended as 'status report' type release, but *real* release with all of the advantages of V3.4.3A should be available soon.   The 'bugs' are so slight that the mods are basically the removal of any 'micro level' compensation for rollof at CD mastering time -- as far as I can tell, there is NO rolloff.  Even a slight increase of +0.375dB at 24kHz for rollof correction almost always decreases quality.

 

Other than this 'micro level' EQ change, the correction for the erroneous HF tilt was completed -- there were some safety limitations in that section of code that was based on the 44.1k/2 Nyquist thing, but since the decoder really cannot work at 44.1k internally anyway (for at least the last couple of years), the restriction was removed so that the max freq is 33kHz.  You might say, but I use the decoder on CDs at 44.1kHz.  That is true, but the decoder runs at 1.5X the input sample rate.   Then upsamples the internal frequency to 4/3 of the internal rate.  Effectively 3/2 * 4/3 ends up being twice the input sample rate.   I did integral sample rate conversions to avoid the 'curve fitting' optimization used in excellent sample rate converters.   The scheme that I use internally is essentially textbook perfect without curve fitting tricks.   The curve fitting on more complex conversions is done to help mitigate distortions before the fancy polyphase filters.

 

My big quandry is:  how do I deal with FA recordings sounding so different from true, pre-FA vinyl?   My decision is to afford the FA style sound balance by default, but make the 'LP' version available with a switch.   Right now, the switch is a bit obtuse: '--hfrl=LP', which means 'HF rolloff LP mode'.   There might end up being a '--vinyl' switch or something similar.

 

 

Link to post
Share on other sites

Bottom line:

Release will have bass that is closer to my own sensibilities -- basically FLAT wrt the gain curve, being continued almost linearly between the 500Hz range on down.   The only trick is in the 35Hz range.   Previously been overly worried about insufficent bass, over-compensating.   All tubby/'overly-full' sounding problems have been resolved.

 

Greater decoding/removal of compression & better calibration (to the 0.1% level of accuracy.)  The calibration precision is so accurate that the apparent signal level is the same at 5 layers or 10 layers.   Not even a 1dB difference as measured based on peaks.   When going below 4 layers, the level comparison (both peaks and audibly.) is difficult because the expansion up to 4 or 5 layers will apparently decrease the average level, even though the basesignal  level is likely within 0.05dB per layer.

 

Correction of the HF problems, being based on NON-FA, true natural quality LPs.

 

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

 

 

Some corrections against the Demos.

1)  I have been getting feedback about 'too much bass'.  Previously, kept getting feedback about too little bass, and I took it to heart.   The next release will have my own sensibilities on bass balance.  Basically, the bass will be 'flat' wrt the response natural within the design.   The only variable that I expect will be the natural rolloff in the 20Hz to 35Hz range, what kind of natural rolloff to use?   Should it be a Q=0.8409 2nd order, or a -6dB at 18.75Hz 1st order?   All I can do is to guess -- the variability of listening devices in that frequency range cause a choice by listening to be very tricky.

 

2)  the decoder has gone through a major re-calibration exercise -- info is in the release notes.

 

3)  With the re-calibration, recordings are found to be 10 layer.   If you take a look at the numbers involved, that actually makes more sense than 7 or 8 layer.   Without the dead-on 0.1% error level calibration, 10layer decoding would produce a greater distortion increase than quality advantage.  At this point, with extremely precise calibration from layer to layer, a 10layer recording maintains almost all of the best quality attributes of an 8 layer decode.

 

Link to post
Share on other sites

Unless there are any new bugs, there will likely be a full release tomorrow.  

Expect the sound improvement to move much more close to a reasonable quality goal than any other release.

 

Some people might suggest that the goal is pretty much met (a few correspondents have already suggested that), but the decoder  needs to be better, and we all really know that.

The demos/Supertramp tests and hopefully enough time for ABBA tests are being run tonight.

 

The release is made reasonable because (I think) the final 'buried vocal' problem has been fixed.   It was the most trivial bug that COULD be fixed, and would have been one command switch for ANY user:  --wof=1.0...   That is IT, that was a big part (not all) of the 'buried vocal problem!!!'.

 

 

The complete release update notes are already ready, but will be posted along with the release.

 

*  This release is the first where I know that a spec can be written (not just a few drawings.)  

 

User capabilities/features relative to the previous full release:

 

* Support for 9 and 10 layers.  It appears that by far, most recordings are done at the 9 layer level, but our current normal 8 layers is usually good enough.   Added -f9 and --f10 switches to enable more complete decoding without needing to remember complicated command.

 

* Faster and improved high quality modes.

 

* The DolbyA section was upgraded to allow very accurate decoding up to 10 layers.   It takes *incredible* precision to be able to decode at 10layers.  It is amazing to me that with the amount and complexity of the processing that the decoder actually produces anything resembling music.

 

* The inter-layer level shift was set accurately to within about 0.05% to 0.1%.   This is important for max quality.   Intelligently setting the calibration is much more meaningful now.  Serious improvements in quality can sometimes be attained by 'tweaking' the --coff setting in sub unit increments.   Normally, still not needed at all.

 

* Per some ACTUAL non-FA recordings, the EQ has now been set to closely  emulate what a recording should TRULY sound like.  The bass is now the only somewhat questionable behavior at this time.   Direct comparisons show very reasonable matches.  (This includes the dynamics.)   The continued bass issue still happens because most/many recordings are rolled off significantly at/below 25Hz anyway.  (A few dB down at 37.5Hz or higher isn't uncommon.)  (The difference in FA and true recording sensitivities have been considered in this claim.)

 

* There is NO 1st order bass EQ between 250Hz down to 25Hz.   The non-2nd order EQed bass IS flat now.   The only variables are two 2nd order EQ -- thats it.  Attempts to attain 'more bass' per earlier expert comments are no longer operative.  Approx same day as one of the previous 'status' releases, a correspondent implied that the bass wasn't quite right -- which means from this person, there IS a problem.   No longer trying to 'make sure' that there is LOTS of bass, not dealing with 'tubby' sound, and vocals less dependent on bass curves.

 

* The decoder normally gives a frequency response that is directed towards FA sound sensibilities.   The old  'tubby' sound even came into being because of the attempt to make the bass strong (per some comments), and within a normally 'flat'/1st order regime, the decoder ended up being as good as possible.   The new 'preFA' switch that provides the character of a real, old fashioned, non-FA recording.  (I had mistakenly called 'preFA' as 'vinyl'.   Of course, the relatively true sound of a decoded recording will NEVER, EVER sound like a current FA vinyl.)  No longer trying to 'make sure' that there is LOTS of bass, not dealing with 'tubby' sound, andvocals less dependent on bass curves.

 

* Improvement in accuracy for AVX512, giving similar precision as double precision mode, but with primarily single precision math.   Accuracy and speed rework throughout the math intensive code.

 

 

Some recent frustration has been the mix of 'bass' and vocal clarity.   The bass matter has had all of the tools to clean it up for a few weeks, but it was tricky to find the two problems once the 'tubby' sound was fixed.   The 'tubby' sound even came into being because of the attempt to make the bass strong (per some comments), and within a normally 'flat'/1st order regime, the decoder ended up being as good as possible.   Essentially, the 'tubby' sound as manifest until a week ago or so was the best possible under the constraint of 1st order EQ only.   'Tubby' bass is now fixed.

 

Once opening the decoder up to another 2nd order EQ, the tubby sound was TRIVIAL to fix, but we still had problems determining 'how much bass' and 'buried vocals'.

 

 

For cases where a specific recording might provide too much bass (a lot of recordings VARY SIGNIFICANTLY on the bass), the decoder software has  tools to do your own mastering.

 

As one of my correspondents mentioned, paraphrased:   there is a point where the EQ must be in control of the USER.

 

 

 

Link to post
Share on other sites
41 minutes ago, John Dyson said:

As one of my correspondents mentioned, paraphrased:   there is a point where the EQ must be in control of the USER.

 

Then this correspondent could be as old-fashioned as per the time we needed bass and treble knobs on our preamps. 

Lush^3-e      Lush^2      Blaxius^2      Ethernet^2     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to post
Share on other sites
6 hours ago, PeterSt said:

 

Then this correspondent could be as old-fashioned as per the time we needed bass and treble knobs on our preamps. 

You mean, back when most of these (the test recordings) were made? 🙂

When they do FA encoding, there really is little need to do mastering, other than sometimes tweaks to make the FA version sound a little better.  FA really bunches up and homogenizes a recording, almost auto-mastering, but getting consistently poor results.

 

OTOH, when the material is passed straight through into the FA encoding, then there has often been NO MASTERING.

 

More recent recordings might be better controlled during the mix.  On the older, difficult test recordings, I am seeing (hearing) vast differences in bass (esp) from recording to recording.   Another  big variant is sibilance.   The Simon & Garfunkel master tapes are well known to be very sibilant on certain songs, and it is very obvious on the decodes.   (Likely shows the lack of mastering.)  This variation in the ancient recordings has become an indicator when the 'bass enhanced EQ' has gone too far.  *On this release, there is little (hopefully none) bass enhancement.   Boosting the bass in any way tends to obscure the vocal. (The method that I used cannot be done with 'tone controls'.), it isn't really a 'boost' in the conventional sense.

 

A lot of the old recordings could benefit from some mastering,  even a WONDERFUL FA recording that has been publically made available here has been mildly compressed.  But that compression was done for a traditional, non-FA recording when the mastering engineer had a lot of control over the product.  The cool thing, modulo my HF damaged hearing, the decoded version comes very close to the true FA LP version, of course, without the mastering differences.   The true non-FA version sounds NOTHING like the normally available FA version.

 

Question:  what should the decoder provide -- something close to FA sensibilities, or an accurate decode?   This is the reason for the --preFA switch (previously misnamed --vinyl.)   It gives the option for decoding into something that is closer to what would be on a non-FA recording.   EQ features are also available in --equalizer mode, where there is no decoding, but the EQ features are available (--pvh, --pvh, --pvdh, --pe,  --lfeq, --hfrl, and now --preFA.)   The '--bh' EQ feature is not needed, because the '--bh' is only used before decoding, but with the '--equalizer' mode, there is no decoding, just pass-through with EQ.  The advantage of --preFA being available in --equalizer mode is that one can do a decoding operation with FA sensibilities, but also post-process to get the original non-FA sound.

 

I do think that we all know that the output decoder EQ isn't finished, I think that I commented on that.

Very regretably, here have been releases done based on my bad hearing also (for HF).   I am trying so very hard to control the experiments, carefully considering the *varying* hearings problems.

 

* THE FINAL EQ IS ESP TROUBLESOME FOR ME -- there are only a few weak 'tells' at the end of the pipeline.   Inside the pipeline, the next step will usually misbehave if given bad data, but at the end, there is NO 'next step', except my bad hearing...  There is only a mild sense of timing mismatch/lack of distortion cancellation on the final EQ. After trying to use whatever weak tells available, the choices must be based on tonal balance. My damaged hearing is not making it easy to finalize the choices. Basing choices on the weak tells only finds an optimization once the general tonal balance has been produced.   It is very, very tedious for me there appears to be a 'perfect' decode when I go to bed for a few hours, then listening after wake up  shreaks at me.

 

This problem with hearing is recently the ONLY reason for release delays.   I am still not sure that I trust my hearing on the 3.5.0? release, so I keep waiting, listening, sometimes trying to compensate, sometimes leaving it alone.  Unfortunately, when dependent on certain aspects of my hearing, it has taken more than the few minutes/hours *that it should* to fix an EQ bug. (I am very comfortable at EE/DSP/CompSci, with the stuff being 2nd nature for 45-50yrs).   It is my hearing intolerably delays the work.

 

Can you imagine the frustration about my hearing?   Fixing the problem is trivial, but because of a wetware problem, diagnosing the problem is so difficult.

 

 

 

 

 

 

 

Link to post
Share on other sites
On 7/1/2021 at 3:20 PM, John Dyson said:

Correction of the HF problems, being based on NON-FA, true natural quality LPs.

This continues to beg the question. Is the goal master tape or LP with its higher level of background/surface noise? Who decides when the decoder matches LP? You, but your HF hearing is admittedly off. 

 

Begging the question because you are asserting that recordings are FA but the "proof" is in the decoding which you are tuning based on your hearing... This has all gotten way too subjective for me. You insist a recording is FA because you can hear it. There is no objective basis for this -- its your subjective hearing. I'm still losing low level HF details in the decodes which are a non-starter. FA decoding does seem to be like a tone control and the best settings or not are entirely subjective.

Custom room treatments for headphone users.

Link to post
Share on other sites

V3.5.1A is ready, and finally passes a few 'tells' that I thought were not possible (including some pre-FA materials!!!)

This is the result of listening/adjustment over several days (different phases of my hearing) until converging on something sane (hopefully.)

No salesmanship is needed for this version.  Even those who might have been overly kind before with pleasant feedback will be even happier.

 

These are OLD recordings that many are not mastered (FA mitigates the need for true mastering in many cases), and also before the time that 'tone controls' were passe.   I did NO mastering on any of the demos except for the correction of the gross LF differences that I note below.

 

The normal demos (everywhere) are ready, and working on some more for specific individuals.

 

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

 

One item of note:  almost all of my recent posts are correct, except one matter -- my claimed difference between 'pre-FA' modes and simple decoded FA are totally wong.   I am getting an almost *perfect* match between legacy materials and the normal decoder output, even to the extent of sibilance same shape/envelope and happening at exactly the same time.   This kind of sibilance can 'travel' and change envelope upon incorrect decoding, but now the only variance is a slight difference in HF (and no vinyl noise, as expected from older materials.

 

* This is great reassurance that the output of the decoder comes very close to the original recordings, but still might be some variance of the frequency response shape.  It *really* is down to tweaks.   There is currently a 'tweak' of about +0.75dB at 21kHz (starts the + curve at about 12kHz), to make the sibilance more correct in very specific cases.    I don't know the source of the rolloff, but the DolbyA units do have a rolloff, and also there could be some rolloff in the recordings.   *This frequency response must be very accurate, or the result becomes amazingly bad.)

 

The big disadvantage to the new version of the decoder -- no 'hearing aid' for us older people with hearing loss.  If there is an incorrect HF boost, should be minor and easy to correct if I know what is going on.

 

The bass also has one known caveat.  A lot of the older recordings might have excess super low bass.   Using the --lfeq=1 or --lfeq=2 (minor LF rolloff) might be very helpful to keep the bass alittle tame.  This is mostly a problem only on certain recordings.   There are a few which need a 'special' --lfeq=99.   That is for only the worst recordings that obviously used a slightly different EQ patterm.   Normal recordings should NOT have an overwhelming bass.

 

The demos (and decoder in a subdir), along with the private things are all available now.   I have attached the 'StartUsing' document for this release.   It is just as easy to download it from the same place, subdirectory of the following demos, along with the decoder binaries. (The 'Start Using' document is still sub-par, but shows the updates and steps needed to make this thing work well.)

 

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

 

 

StartUsage-V3.5.1A.pdf

Link to post
Share on other sites

A very, very minor update...

Once doing the V3.5.1A release, I started reviewing some minor issues. I turn down easy projects like other decoders, simple compressors/DAW stuff.     The only real problem in the last several months has been about bad hearing and that the LAST EQ producing only weak or non-existent 'tells'.  

 

This decoder is for those totally disgusted by the quality of FA recordings that are normally sold.   This is not for those who have adapted, because the decoder IS cumbersome to use, and will forever have flaws, however the flaws will also be diminishing.    For some people, the decoder IS a godsend, for others -- who cares, many others might really prefer the hissy, dynamics distorted, strange high end, swishy cymbals of the FA sound..   Anyone with the interest, technical background and good hearing could have already produced the final EQ themselves.   The technical side of this is TRIVIAL for an engineer with reasonably good and reliable hearing..   It is impossible for those with bad hearing.   A competent engineer with bad hearing doesn't benefit from his/her competency.

 

Hopefully V3.5.1A and V3.5.1B will alleviate most of the lingering problems, with V3.5.1B doing even better.   The only way that the complex internals of the decoder are functional at all is that I am persistent, enjoy tough challenges, and pretty competent.   As long as several days are spent on doing a release, instead of the hour that should be, most of the dependencies on hearing can mostly be mitgated.  (I did start learning some LF 'tells', but they are very weak.)

 

1)  Why are --f10 decodes so problematical (mentioned in the documents)?   Answer just found:  I let the attack/release control signal get ahead of the audio.   The V3.5.1B will have that fixed, but makes only minor difference on V3.5.1A.   This fix will allow full decoding of some of the best recordings.  --fa almost always works 'good enough', certainly better than the original recording (nowadays), but that extra detail can now be available.  '--f9' is icing on the cake, and when used properly, '--f10' can be yummy icing on the cake.   Frankly, I could just barely hear the '--f10' problem on certain, very limited recordings, so it is a knife edge problem, and it is the uncommon nature of the problem that temporarily forced it on the back burner.   I did NOT think that the problem would be so easy to solve, until I solved it 10 minutes after the release.   By making the problems with '--f10' go away, that certainly might make improvements to --fa and more importantly, additionally --f9.

 

2)  A bit too much bass...   The bass thing is very complex, and even though it is simplified compared to what it used to be, there are still some strange things that the inventors of FA did to the signal.   This actually makes the bass more tricky than the midrange, which is trickier than the HF (other than the pre-emph/de-emphasis.)   It is okay to user to apply  some rolloff (e.g. --lfeq=1) until the new release is available.   In V3.5.1A, there is about a 1dB bump at 50Hz and 1dB bump at 25Hz.   The fix will mitigate those.  (The 'bump' isn't really a 'bump', but instead is a signal that is 1dB higher than it should be.   There is actually a gain decrease at those frequencies, but the decrease is just a little too slow.)  As a complication:  Some of the 'bump' is needed, just how much?

 

There is little reason to wait for V3.5.1B, but it will be coming in hours, not days.   The demos need to be updated.   I will NOT be updating the 'Start Using' document because nothing has changed for that.

 

I REALLY didn't think that these would be so easy to fix, and since they are so easy, I thought that it would be a good idea to just fix them.   After this, I am gone for probably more than just a few days.   At this point, the decoder should be 99% of what would be expected of it*, especially after this last fix coming in a few hours.   The decoder itself is ready, but must do at least the standard demos before doing the release (as a matter of responsibility, must stay away from listening to recordings until the demos are ready -- then use the best hearing that I can muster to review the response balance/etc.)   The small amount of distortions in V3.5.1A have been obliterated in V3.5.1B, and I didn't even know it was possible.  This is almost like the total correction of the join between MF and LF so that some of my old 'impossible' tells are now corrected!!!!

 

*95% of the decoder has been correct for the last 6months, it is just the *simple* EQ that my unreliable hearing cannot judge.

 

What a real relief.   The only question now is the HF and LF balance.  I cannot judge that.

 

 

 

 

Link to post
Share on other sites

V3.5.1B is available *FINALLY* with "hard of hearing" enhancement removed!!!!

 

1) Hard of hearing enhancement removed, which oddly matches one of the standard EQ building blocks.  (18kHz, -6dB).

2) Bass now based on a totally flat (not partially flat) EQ.  The mistake was my massive 1st order EQ at 75/6 and 75/8Hz.   Removed that, everything fell out to two 2nd order EQ added to the signal coming from the MF to LF adaptor.  (It isn't just a normal EQ, but has a cross-hatch pattern to match the signals.)   Still have some 'speaker code' safety code. Will explain in document.

3) The --f10 is fixed along with --f9 and -fa improved in clarity (--f9 is improved more than --fa.)

 

The standard demos along with decoder binaries are available, still uploading some of the other stuff.   Wont' be fully updated for several hours. 

 

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

 

 

* Found my hearing doesn't diminish if I lay back at about 45deg angle.   Just as I am typing this, I can hear my hearing diminish without any indication.  I now have a good basis of the missing hearing.  Don't know why I couldn't hear it before (other than adaptation/accommodation.)

 

Link to post
Share on other sites
On 7/2/2021 at 5:16 PM, jabbr said:

This continues to beg the question. Is the goal master tape or LP with its higher level of background/surface noise? Who decides when the decoder matches LP? You, but your HF hearing is admittedly off. 

 

Begging the question because you are asserting that recordings are FA but the "proof" is in the decoding which you are tuning based on your hearing... This has all gotten way too subjective for me. You insist a recording is FA because you can hear it. There is no objective basis for this -- its your subjective hearing. I'm still losing low level HF details in the decodes which are a non-starter. FA decoding does seem to be like a tone control and the best settings or not are entirely subjective.

 

* First, with the 'hard of hearing' enhancement being removed in V3.5.1B, the comparison should be easier.

 

* Second, as commented on several times below, I will NEVER even discuss whether or not someone should prefer the decoded or original versions -- I don't believe in telling ANYONE about what their taste *should* be.   Even for myself, based on my bad/variable hearing, my choices might be different than someone who is a true golden ears.  My hearing/choices become important for the project ONLY because of the project.  The hearing problem is a very common disability for many people.

 

Originally, I wasn't going to respond because the discussion could be couched in 'taste' and 'preference'.   But apparently, it will require true 'existence proof' to be able to show the difference between compressed, dynamics distorted FA and the original character of the recording.  AGAIN: I am NOT dismissing your (or anyones) preference, just requesting an open mind about *accuracy* and similarity.  Accuracy has only a very approximate correlation with preference.

 

All I can say is to compare the decoded version of 'Lamb' with the pointer below with the LP version posted earlier  (imagine the ticks and pops removed.)   Listen carefully for the tells, when you can hear the tinkles start, etc.   You'll find, except with a little compression, very little, on the LP, the sound is identical.   The decoded versus ORIIGNAL versions still sound a little different because of the likely 30+yr old vinyl and mastering, but otherwise the sound is pretty much identical.   The LP was likely mastered, but most FA recordings don't appear to be mastered before FA encoding.  In fact, just checked some MFSL recordings, and they appear to sometimes 'master' based on the FA version to start with!?!?! (An example of the lack of mastering on FA recordings, decode/listen to early versions of Simon and Garfunkel FA recordings, you can re-create the sibilance that is well known to be on their master tapes!!!)

 

* I will remove this link in about 7days.   If you need a copy of the original LP, and the previous link goes away, I can make it available PRIVATELY -- just tell me that you want a copy for comparison purposes.   (The decoded version is three seconds longer at the start, so the timing will be three seconds different between the LP/original version and the decoded versions.)   I check for every nuance (except with less accuracy above about 7-9kHz when my hearing is broken.)

 

https://www.dropbox.com/s/hm3z6or09xu0v8l/01 - The Lamb Lies Down on Broadway-DEC.flac?dl=0

 

On the other hand,  the distributed FA version has a pronounced increase in HF ambience, even though the signal level compression at midrange is nil.  The average (MF) compression is NOT the problem with FA as much the HF compression (squishy cymbals, and weak, smeary sibilance.) * Also, there is a LOT of bass dynamics distortion in FA.   The false HF ambience is one of many side effects of HF dynamics distortion.   The HF compression effect allows hearing the 'tinkles' and the fly buzz at lower levels.  No matter what, the resulting compression changes the recording from the original.*   In a way, FA encoding is 'automatic mastering', but, in my opinion, FA is mastering done very poorly.

 

*Watch the running log of the gain '--info=2', and note that the MF (midrange) seldom does any gain control, and probably is the reason why the MF dosn't show much compression or dynamics distortion.   Practically ALL of the FA dynamics distortion/compression happens in the 20Hz->180Hz range and the 2kHz->20+kHz range.  The gain changes in the HF range is sometimes on the order of 40dB up and down several times per second or faster.   The gain changes in the LF are often much slower (in the range of 5-10dB/second at most.)  Often, there is little LF gain change, and almost never MF gain change.

 

Which do you prefer? I won't argue about that.  I REALLY, TRULY WANT you to enjoy the version that you like.   In fact, good compression, artfully done, can definitely improve a recording.  I dare say that the LP, with about 3-5dB of well done LF/MF/HF compression is likely an improvement over the master tape.    Some might say that in some instances, the increased HF ambience of FA might improve some recordings, but that is NOT what the project is about.

 

The decoder is simply about attempting, more and more succeeding, to return the recording to the original.  Even though I resisted talking about 'Lamb' specifically for about 1wk, the decoder comes fairly close to the original, without the 3-5dB of proper low level compression and  'wiithout the expected aging of the vinyl'.   On the original (LP) vs. decoded versions, the vocal sounds/mild sibilance appears identical at about 40-45 seconds.   If decoding incorrectly, all kinds of h*ell breaks loose on that kind of transition.

 

With the undesired and unintentional  'hard of hearing' 'house EQ' being removed,  it is probably factual to say that the decoder comes fairly close to what was originally by the devils in the FA encoding machine!!! :-).

 

 

 

 

Link to post
Share on other sites

John, did you ever get your hearing checked out? From the sound of it you might just benefit from a maneuver that clears your inner ear from debris. 

ISP, cat8.1, Zylix switch on LPS, Finisar <1321>Solarflare 8522, AMD 9, Aorus X570 Master, Corsair AX1200i, 16Gb Apacer Ram, Pink Faun I2S OCXO on LPS, home grown RJ45 I2S cable, Metrum Amethyst (modded to accept I2S), Klangfilm 204a, Klangfilm Trionor (3Tesla 835, JBL2402)

Link to post
Share on other sites
3 hours ago, John Dyson said:

The decoder is simply about attempting, more and more succeeding, to return the recording to the original.  Even though I resisted talking about 'Lamb' specifically for about 1wk, the decoder comes fairly close to the original, without the 3-5dB of proper low level compression and  'wiithout the expected aging of the vinyl'.

 

John, pardon me ... I am trying to read texts like these with the required interpretation hence context of your hearing. ... Even if I read this 20 times, there can't be truth in what you are saying. If you can not hear well, you can not hear well, and you would not be able to judge this (see quote) just the same. This also comprises the decoder being perfect for "years" and as many times as days exists in those years.

I hope you don't need these texts for your own motivation because IMHO it really is not useful.

 

Just saying (and derived from your conversation with Jonathan - not really with me because I learned how to deal with it (I think)):

 

Generally, in parallel two streams exist:

1. The user making (IMO) useful comments, like no running water in whatever song to be heard, while there should;

2. You going your own way after (!!) such message, taking out the reference of that message, possibly claiming that no versions will be lost because you save them all (which I believe). Later, you may come back at the issue (like you do here with Lamb) but nobody has a reference any more because you are thee+ versions further.

 

It all reads like a trick (read : sales trick while you sure are not trying to sell anything).

 

If you ask for help, again IMHO, you should react to that before proceeding with your own ways.

The best example could be me myself (and you). The responses are countless, but in the end are useless for myself because in no circumstance that I recognize, you attacked the issue and a new version accompanied with a "so Peter, did this help ?!?" ever is in order. What does happen all the time is a "OMG, again all changed !" from my side. Do you realize this ?

 

Somehow I started this post with telling that you can't know yourself what "is" correct. You should know by now because of the number of days in those years, etc., right ?

Mind you, I work quite similar with my own customers. If day tell me that running water has disappeared, by response would definitely be "Okaay, then we'll get that back for you". But I could also say "and what about that bass then ? is that for the better now ?".

John, I am not telling you how to live your life, but it is you who ask for help. In that case to some extent those who provide the help have some small rules too. They don't want to see their help debunked. And I know, you are not doing that explicitly. But doing it implicitly brings not so much difference ... (avoid it and proceed with your own thing).

 

PS: If I had something to say about the last 2 weeks, I would say that you should immediately cut out the sudden LP stuff. It is not related by miles, LP sucks anyway, and apparently it creates new decoder or EQ version and those who helped have to start over again ? WHY ? who starts the LP sh*t ? It was mentioned. This does not really mean that now all should sound like LP. For heaven's sake, I hope not.

 

PPS: I am not suddenly over a hurdle with this post. No problem. But I read the "all is perfect now" one time too many; you cannot hear that and with that can not know that. You present it as fact ...

OK ?

Lush^3-e      Lush^2      Blaxius^2      Ethernet^2     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to post
Share on other sites
3 hours ago, MarcelNL said:

John, did you ever get your hearing checked out? From the sound of it you might just benefit from a maneuver that clears your inner ear from debris. 

I have been careful to keep my external ears VERY VERY clean.  I have been perfect on hearing hygeine esp the last few years (for obvious reasons.)   But the inner ear just might have a problem.   Since my hearing is different vs. position, and is always better first thing in the morning (I sleep at a 45deg angle), then within about 15-20minutes, it goes to hell.  I just figured out a part of the pattern a few weeks ago.  Now, I found that it is purely positional, and SEEMS like fluid buildup.  (Now, since I know what to look for, during this typing, I can hear my hearing diminishing RIGHT NOW.) 

 

It DOES appear that there is a fluid buildup of some kind, but it is bilateral...   This might mean that there might be a sinus problem or somesuch.   I found that once I sit at the same 45deg angle that I sleep, after about 1/2 hour, my hearing is cleared up.

 

The hearing loss becomes so profound that the recordings sound like they have a severe bass boost, with inadequate HF.   The inverse of that should sound similar to the demos (and decoders) that I provided previously.   That is, the recordings had weak bass, and too much HF...

 

At least, the problem is better understood, and I have a workaround, however imperfect it is.

From now on, my listening will be happening at a 45deg angle!!!

 

John

 

* about 1Hr ago, I paniced when I thought that I had uploaded the wrong demo versions.   Well, that was wrong, so I am restoring the original demos that I provided several hours before.   I have a faulty way of handling the generated flac files, and when listening earlier on, I was listening to the last V3.4.X version, which, of course, shreeked.   Once, after careful review, I found that I was listening to the wrong version, and tried the raw files, I realized that the raw versions were indeed correct.  (The raw versions are never comingled, but the .flac versions are.)   I just did a wildcard on the directory when listening.  On that wildcard, the shell grabbed the worst one first!!!)   Can you imagine my panic?

 

 

Link to post
Share on other sites

This is about how the V3.5.1B bass and HF EQ was changed on V3.5.1B...

 

The bass 1st order EQ  now totally flat coming out of the MF/LF EQ bridge.  (That 'bridge' is not normal EQ, but a strange and complex merge of EQ with frequency offsets.)   Below that 'merge', the 1st order EQ is totally flat to below 25Hz.  (25Hz is one of those magic frequencies.)  There is still the additional 2nd order EQ at 75Hz (+3dB at 77.5Hz, +3dB at 75Hz and +3dB at 72.5Hz, each with Q=2.0).    There is another 2nd order EQ at 50Hz, (+3dB at 52.50Hz, +3dB at 50Hz, +3dB at 47.25Hz), also Q=2.0.   There is also a complimentary EQ at below audio to bring the gain back down to sanity.

 

The interesting thing about these 2nd order EQ, the need was identical whether or not my hearing was messed up.   Those shows that the 'tells' were correct, but there was a severe bias in the way that I heard.

 

(Previously, the complex curve probably showed the inverse frequency response of my hearing, when in a damaged state, at LF.)

 

I have been trying to figure out why so much EQ.  Perhaps an attempt to subvert some of the LF compression to avoid some of the rather significant LF distortion caused by DolbyA units without decoding the DolbyA compressed signal.

 

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

 

The HF region was a little different, less based on 'common sense', moreso on 'engineering sense':

 

The pre-emphasis/de-emphasis confused the additional needed post-decoding EQ.   When I did the EQ that 'made sense' from an EE/DSP standpoint, then the result didn't have enough HF per my hearing.  So, I thought, well maybe my concept on the necessaray post-decoding EQ was wrong.   Therefore, I pulled away one of the 18kHz -6dB EQ stages.  Even further back, things were all confused because I was thinking that the pre-emphasis/de-emphasis also needed to be modified away from my original conjecture based on my engineering background.

 

Bottom line:  If I had trusted my engineering know-how and my very precise understanding of what DolbyA units do -- the sound would have been much closer to correct, rather than vetoing the engineering know-how by depending on my strange hearing.   Of course, any honest person whose ego allows for mistake, would do what I did (trusting the obvious, even if it was wrong.)   It might have been better to be an overly self-assured and arrogant engineer, because in that case, HF would have been closer to correct.

 

Honestly, I still have a caveat about the HFeq.   I am not sure if it still needs an 18kHz, -3dB EQ.   The reason for this is that there is a well-considered pattern in the pre-emphasis/de-emphasis curves, and it seems like it might be plausible that one additional 18kHz, -3dB step MIGHT be needed.   This might be confused by the needed 24kHz, -6dB pre-emphasis and 24kHz, +6dB de-emphasis pairs being above the hearing range, and there was consideration by the original designer.

 

This 'reveloation has a bit of a relief that maybe I'll be able to better maintain the program once it is fully complete.  (There are still several significant performance enhancements in the queue.)  I might even look into doing a GUI.

 

 

 

 

Link to post
Share on other sites

This is about the decoder project, NOT  my hearing -- but I am working on resolving the problems.

 

Remember in the previous post where I had predicted from an engineering standpoint that additional EQ is likely needed?  (I gave the reasoning in that earlier post.)   Well, that 'prediction' was true.  During an experiment right after the previous post -- I found that  my hearing diminishes within about 1-2minutes, and then starts getting worse in the 5-10minute timeframe.


By the time that I decided not to add the additional EQ, my hearing was already damaged.   The exact EQ will be figured out within about 1/2Hr, and will have the decoder built within 5 minutes after that.  THEN, it is a decoding session.  It does seem like 6hrs should be an easy goal.!!!

 

I GREATLY APOLOGIZE, but the next release will have the (hopefully last) hearing correction.  Geesh, I had thought that the previous was the correct correction...   It will be triival to get the release out by about +6Hrs, and that should include the private examples!!!

 

Maybe the industry pushback might become more severe?  :-).   Look for the money trail :-).

 

John

 

Link to post
Share on other sites

V3.5.1C has been available for quite a whiile.   I have done EVERYTHING that I can to keep the EQ configuration choices close to my original 2nd gessing of the orignal designers, based also on DolbyA behaviors.  (that is, using 2nd guessing based on my own engineering/EE/DSP background.)

 

I didn't' announce so that there was some time to double check and pull the versions back.

The files are in the usual places.

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

 

The HF EQ choices are very very similar to my original intuited concepts (before my hearing got involved,  panicing me into bad decisions).   The results MUCH less depend on my hearing, but did have some luck with my 'rested' hearing agreeing with the results.  Changing from he old shreaky stuff to the current stuff implemented by adding back in some 9k and 18k -3 or -6dB EQ, ending up pretty close to my first engineering estimates.   There is some minor variance from some of my original guess as had learned some things.  (The varitions between 'correct' and 'shreaking' is a matter of exclusion -- not totally different EQ.)  * the  'hearing aid' version was similar to my original design, but with lots of -dB EQ removed to 'sound right' -- BAD decison.

 

* The one-off demos will be updated ASAP (minutes, not hours.)

 

The LF might still be troublesome, but spent lots of time making sure that the design is coherent yet producing reasonable results.  Until a few hours ago, with the EQ being very compatible with most recordings, I didn't realize that Supertramp recordings were EQed differently than most other recordings.   This means that my attempts to make 'Crime' sound like the FA version, while making other recordings sound correct also -- FUTILE.   I had been fighting that problem for months, assuming that I didn't understand some kind of subtle magic.   The 'subtle magic' was the way that the Supertramp recordings were prepared.    During decoding the Supertramp recordings with this hopefully more EQ-correct decoder, I have found inconsistancies between the group of 4 Supertramp albums vs most of the other recordings.

 

* THE TOTALLY IMPOSSIBLY SIMPLE, TOY, single layer DolbyA modified decoders could be tuned for each recording.   Even though the program didn't really do enough to 'decode', it coulld be made to sound okay -- but for only one recording.  Each recording needed 5-10 different EQs, manually applied.  The new decoder requires VERY LITTLE tweaking.   The 'setup' on the old, play-toy decoder required a redo for practically every recording -- basically useless.  THAT is why I started developing an infinitely more complex, but more complete decoder.

 

Both versions of the demos are available, and I have others of the demos (e.g. stuff that the active reviewers check out) are completed, but on hold because my desire to do QC and listener fatigue

 

As mentioned above, I had mistakenly used Supertramp's recordings as one of the references for the EQ.   This EQ/mastering might be one reason why the recordings might sound good to some audiophiles.  That was a very bad decision for me to focus on Supertramp during testing, beause the Supertramp FA recordings have been mastered.   This means that there is EQ that must be undone after decoding.  The EQ is minor, but is a very important issue.  Just because audiophiles sometimes like to use Supertramp for reference does NOT mean that the FA versions were good for my purposes.

 

My hearing has started to close down for today, even with my uncomfortable -45degres working position.  It is so odd to have gotten beautful results earlier in the day, then wonder why the results are so dull now.   If my hearing returns today, I'll work on more private demos.

 

 

Link to post
Share on other sites

Most fun -- the next time that I do a decoder release, the demos will be decoded at --f10, when appropriate.  This *really* brings out the details, long since obscured by FA and even lower levels of decoding.   Since I just had this observation about '--f10', I'll be doing these full quality, much more clean demos very soon (maybe even today.)  THIS IS REALLY WORTWHILE, but slightly slower to decode.

 

 

There is a very subtle bug in the decodeer where the interleaved frequency offsets weren't correct for two layers of the LF post decoding EQ.   This causes a minor sense of LF almost 'rumbling' distortion -- probably very, very unnoticeable for most people, esp on speakers.   Also, going to revisit the EQ the next time that I can -- probably a few days.  There have already been a few very minor mods to the frequency interleave in the EQ, but again, mostly very minor tweaks.  (This interleave is likely unknown to most audio hobbyists, and is related to the effects of dynamics processing and maintaing a smooth sound when the FA signal is so severely mangled.)  There is benefit in the interleave, and I'd like to know how someone figured it out.   Of course, I found the interesting design characteristic once the decoder *dynamics* became close to accurate.  I AM amazed at the genius,

 

Be very assured, the decoded results are coming close to the actual  mix, not compressed in such an abusrd way any more. (in the dynamics sense, the decoder has been DEAD-ON for months), the final EQ has been the trouble.    I was thinking about cases where people not noticing FA -- probably cannot hear it on most speakers, and decoded material WILL sound a little less 'bright'.   Need to match the unrealistic FA brightness  is why the partial layer of final EQ will be normally disabled in the next release.  This will return the boosted FA brightness.   The 'preFA' switch will be repurposed to add the technically correct layer back in.  (Disabling the layer amounts to about +1.5dB at 18kHz.)

 

It was very interesting to hear the recordings as they are (without severe hearing problems), and noted interesting details that I had not recently heard.  The 1st order EQ is also a real 'hoot' about making sure that the EQ is correct.  I would have never guessed about the subtle/not-so-subtle high frequency effects of incorrect 1st order EQ at nomially lower frequencies.  Oddly, and I am not sure why, incorrect 1st order EQ in the LF can have profound effect on the sound of the higher freqs.   If the lower freq EQ is not correct, it appears that the higher freqs can be partially cancelled.  Some day, will try to figure out if it is related to phase or related to delay.    THIS WIDE RANGING EFFECT does NOT happen with PARAMETRIC, GRAPHIC or normal 'Tone Control'/common 2nd order shelving.  This interesting effect only appears to happen with 1st order EQ.  * A GOOD side-effect of the 1st order behavior is that once the EQ is close to accurate, the 'tells' from 1st order EQ are good hints to zeroing in on the correct 'blocks' in the decoding EQ.

 

As it is now, when listening with 'good' hearing, the recordings appear to be technically clean, but might benefit from the removal of the one layer mentioned above (+1.5dB at 18kHz.)   The frequency response of the decoding is much beyond 20kHz, and probably reaches 30+kHz, even though the CDs are brickwalled at 20.5kHz or so.

 

I will be trying to enjoy the output of the decoder, laying in the unpleasant 45deg angle, but also listening for defects.   There might still be a bass issue, and the Supertramp recordings are *interesting* to say the least.   I will do further review on those, just as I will on other recordings that appear to need post-decoding EQ modifications.   This need for post-decoding EQ modifications in the Supertramp recordings is indicative of 'mastering' to some degree. I just wish that I hadn't previously depended on Supertramp as an EQ reference...  What a waste of time!!!

 

 

 

 

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now


×
×
  • Create New...