Good status of the recording correction (FA) project

[John Dyson]

Here are the new hand-in-glove decoding results, with the HF correction and consideration for the bass

I did refer to the Beatles remaster CDs, and the bass results from decoding are very similar, even though there is too much bass on BOTH (IMO), the decoder is doing the correct thing. On the other hand, the Tijuana brass results were relatively bass heavy, and there is a decoding option to correct that matter (--fa=LL), and it does sound better. Unfortunately, the bass EQ change did not include a change in the bass pre-emphasis. It sounds like the bass pre-emphasis needs to be disabled, but I need more measurement and evaluation to determine what is needed on those specific recordings. Therefore, I suggest caveat on 'Casino Royale' and the other Tijuana brass recording.

Before and after examples:
https://www.dropbox.com/sh/tepjnd01xawzscv/AAB08KiAo8IRtYiUXSHRwLMla?dl=0

 

[John Dyson]

No, what I heard is this:

Robinson:
View attachment 122385


Sun:
View attachment 122386

Mamma Mia:
View attachment 122387

Danny:
View attachment 122388

I think we can stop here... the exact same pattern on all four files.

In general the tonal balance is totally screwed, we're talking a scary peak/valley landscape of +8dB/-10dB of EQ changes applied in the 20Hz to 20Khz range! No wonder it all sounds honky and zingy. Do you ever sanity check your data? Obviously you don't.

Honestly, I'm out here. You're either a troll, or living in a bubble, or even worse, made up a story in order to spread Trojan Horses.

Here are objective averaged spectral measurements for the decoder: TOP is the RAW CD, Middle is the decoder, Bottom is vinyl. It appears that the vinyl is at variance, but the CD and decoded versions are similar (as expected), and the decoded version does show some expansion downward as expected. Appears that you show biased and unreliable data -- that does not bode well for your credibility. Also, I have other honest evidence from others.
Per these totally objective (not subjective) measurements, there is not really 'TOO MUCH BASS' or other such complaints. Because these are averaged spectra, the exact details are not applicable, but the general shapes are. The greater bass on the CD comes from bass EQ
manipulation to make the RAW CD sound relatively normal under compression. Likewise, the flat LF spectrum on the vinyl is expected (density should roll off, so appear flat.) The decoded version shows a similar curve to the CD, except rolls of at inaudible frequencies to protect speaker cones/etc. (I actually consider those issues.) We have determined that there is HF EQ on the vinyl oopy (it came from an source without provenance anyway -- just using for quick reference.) The CD is what was used to decode the decoded version though.

Because of the unscrambling below -30dB, ONLY the general shapes can be compared -- exact numbers are meaningless. The highest (peak) signal levels are the most helpful, but the average spectrum only shows a guideline.

 

[KSTR]

After undoing the EQ I get a RMS null about -50dB deep and a roughly corresponding linearity plot:

Your SW seemingly does nothing else than add some static EQ, no sight of changed dynamics at low levels.

I understand your plight and limitations, but it is really best to be kind to people, even the disabled. So, I will do so.

Uh-huh. I'll do likewise, Have a nice day, Sir.

 

[John Dyson]

After undoing the EQ I get a RMS null about -50dB deep and a roughly corresponding linearity plot:
View attachment 122394

Your SW seemingly does nothing else than add some static EQ, no sight of changed dynamics at low levels.

Uh-huh. I'll do likewise, Have a nice day, Sir.

I thought that you were done communicating with me -- so much for integrity.

The dynamics processing is aggressive -- I'll show a movie in the snippets directory later on.
The same dynamics processing has been accepted by professionals as a relatively accurate DolbyA decoder (unlike the Satin plug-in.)

Me thinks that you are in over your depth...

 

[John Dyson]

Good news. I decided to do another release. V2.2.3C. I have also added some screen-grab movies (at the bottom of this posting) of some snippet decoding sessions.

https://www.dropbox.com/sh/1srzzih0qoi1k4l/AAAMNIQ47AzBe1TubxJutJADa?dl=0

This fix is all about the LF EQ, and how hard it is to do correctly. The big problem is that it is necessary to dispose of 30dB of signal between approx 500Hz and 20Hz (where 20Hz is approx -30dB). A straight EQ curve is NOT correct. Also, unlike HF, there are no good 'tells' that I can discrern, so it is all about developing prospective engineering rules, and perceived response balance. My own design style is that I do NOT depend on my own listening taste and limit myself to using two things for my input: Other, honest people's suggestions, and the 'rules' that I have laid out based upon my actually rather deep engineering (EE, DSP, CS) knowledge. I even listen to people who I don't really trust - just try to be more careful. Success comes from learning and having an open, engineering and somewhat scientific mind.

The people so kind to test and review the results, they previously kept asking for more bass. My own headphones are nearly flat to below 20Hz, so excess bass is actually painful for me. I don't have a good view of what bass should be (also hearing problems), so I must depend on external information sources. Part of this bass problem though was a typeo in the code -- so things got mixed up at the last moment.

I actually reviewed what my local skeptical nemesis/friend has said, and my thinking agrees about too much bass, but I am pretty sure it is different than what he suggests. Basically, in some naysyers minds, the entire decoder is wrong (their competency then falls into question), but there are and HAVE BEEN bugs. The most difficult bug for me has been the LF EQ. Unfortunately, along with the LF EQ there is LF pre and de emphasis, which complicates matters very significantly. I strongly believe that listening to people, however painful, is important -- and I do learn and almost never ignore.

Bottom line: I have decided (through various measures), to trim a max of about 3-4.5dB starting at about 30Hz downward. Amazingly, this makes a huge difference to improve listenability. I felt that I could do this change because I have re-interpreted my rules. In one case, if I do compensatory EQ like the necessary 80Hz, Q=1, +6dB boost, then there must be a corresponding decrease further donw in the spectrum. Before, I was using 15Hz because I wanted to get 'all of the good bass'. That was obviously wrong, so I did the intuitive move the EQ to 20Hz to wrap up the 80Hz LF boost. Additionally, I have historically matched a 25Hz -3dB 1st order rolloff to an additional 20Hz -3dB 1st order rolloff. Why did I do that? It is a matter of one of the few 'tells' in the LF region. I rescended that rule, removing the 25Hz rolloff to keep 'all of that beloved bass'. That was also wrong to move the 20Hz compensating EQ to 15Hz, so I have moved it back to 20Hz. Since the Q=1, there is some effect away from the 20Hz range, and that is good.

Bottom line -- KEEP FOLLOWING YOUR ENGINEERING RULES, stay away from subjective design decisions -- they will often lead you astray as they do me.

Here is the new V2.2.C release with the bass that even I MUCH prefer (the doc wasn't updated as there is little to say):

https://www.dropbox.com/sh/1srzzih0qoi1k4l/AAAMNIQ47AzBe1TubxJutJADa?dl=0

Screen-grab movies of some decoding sessions. Even chose a more contemporary 'Call me, Maybe' by Carly Rae Jepsen. This 'scrambling' problem in consumer recordings AS DISTRIBUTED persists till at least today... I can do any number of recordings if someone needs proof, even willing to try to find a way to do it realtime for over-internet viewing.

https://www.dropbox.com/s/5dcqiab0fhnirxx/ABBA-Bobby-2021-04-06_13.16.31.mkv?dl=0
https://www.dropbox.com/s/d0vny81zskjpf3x/Carly-Jepsen-2021-04-06_13.10.09.mkv?dl=0
https://www.dropbox.com/s/5n66rhxh2zsxjwo/Frida0-2021-04-06_13.14.42.mkv?dl=0

VERY IMPORTANT: in the movies above, the running gains are NOT synchronized with the audio. There are 'seconds' of time delay through the decoder, and each layer has delay. Of the 'three' gains for each frequency band, look at the first one -- that varies more in relation to the loudness of the recording. Note also that the pairs of gains for each layer/each band are MID SIDE and not LEFT RIGHT. If you run the program in DolbyA mode, it works in 'LEFT/RIGHT' mode.

Also, you should notice that sometimes (often in some recordings ,never in others), there might be three or four bands swinging 10 or 15dB per second sample, and if you know the internals, that 'swing' can happen about 5 times between samples. So, the gains change A LOT, perhaps 300dB back and forth all togehter for the 9kHz frequency band... The decoding is REALLY very dynamic. However, it doesn't happen much at levels -20dB or up, ALL of the decoding is about de-scrambling the lower levels of signal. The design of the detector and gain control allows it to dig DEEP into the signal to do its clean-up.

Do NOT expect the sound of the expander complex to sound like a simple expander in the production studio -- this thing is exquisitely complex and has a purpose different than a normal expander, but DOES use expansion elements.

 

[KSTR]

I found some time to play with the latest revision, which is V2.2.3E btw, as contained in the archive linked above.

I used the Anne Murray original CD track snippet "Danny's Song" as provided (because it is an excellent and clean production) and applied the decoder, using a batch file:

da-win.exe --info=11 --fz=opt --fw=classical --fa --input="%1" --outgain=0 --floatout --overwrite --output="%2"

After that, I downsampled it from 88.2kHz back to the original 44.1kHz with Adobe Audition to make things easier (could have used SoX as well but all this was done on a Windows machine).

Again, and as expected, the result showed Mr.Dyson's hard-coded personal taste EQ, the same curve we've seen before and which is probably present in any of his examples:

The frequency response was obtained by placing a (REW-generated) LogSweep at the start of the track which then was converted to an Impulse Response with REW.
An interesting detail is that the phase does not match the magnitude response at the kinks seen at 3kHz and 9Khz and the irregularities at ~1.5kHz, this suggests the linear transfer function of the decoder is not minimum phase (if it were min-phase the wiggles would be fully reflected in the phase response as well, can be checked by letting REW compute a min-phase version).

This impulse response was used (I) to de-embed the EQ function in the decoded track in order to restore its original frequency response for a meaningful listening comparison, and (II) to apply that same exact EQ to the original so that a full-fleshed subtractive analysis is possible with a result that can be actually listened to (as DeltaWave's built-in FR compensation produces unavoidable processing artifacts).

The corrected FR for listening was created by REW, generating filter parameters, which then where converted to an impulse response with RePhase which in turn was used as a convolution kernel in Adobe Audition. This is a minimum-phase correction so the (small) excess phase contribution is preserved. The corrected FR looks like this, with a tolerance band of 1dB (I didn't bother to correct stuff below 40Hz as there is little content anyway).

I could have used my double-precision time-domain convolver as well but the ultimate precision and complete absence of artifacts is not needed here. But I used the convolver to obtain the "linear-transfer-function-equalized" files to feed DeltaWave, for a clean residual. I only used the left channel data for this.

The linarity plot obtained from DW shows a 1:1 transfer for the first 10bits (-40dB) and below that the gain decreases, showing a slight expansion. The plot is not really representative for what's going on because, as explained by Mr.Dyson, there is a bit more going on that just a simple single-band downward dynamic expander, but it shows the tendency:

More revealing is to actually listen to the residual, where we can find what the decoder "took away" from the original. We can hear that it is mostly recoding noise, reverb tails and general lower level content, whereas the loud voice is about equal in both (and therefore mostly cancels). The residual also has a certain roughness and pumping to it, exposing the dynamic processing.

This also what one can hear (when turning up the volume) in the direct comparison of the original and frequency-corrected "decoded" version. Personally, I'm not sure if I like the processed version, it sounds quite dull in the quiet sections (on the guitar, notably) and all the ambience details that make that (rather clean) recording sound any natural/live are blurred or outright missing, the downward expansion is just too drastic. Stereo image I don't find to be improved, rather the contrary, by this.
The differences are large enough that most people will be able to succesfully ABX this (if anyone feels a need to do so)... just turn up the volume quite a bit.

Link to a ZIP archive containing the following files:
- Original snippet ("raw").
- "Decoded" snippet as per above settings, with restored frequency response as explained.
- Delta (residual file), with +40dB boost to make it audible.

My conclusion. Mr.Dyson's "Decoder" actually does something (besides that nasty gross EQ which basically spoils its usage for non-tinkerers) and it might actually improve some tracks or find its place as a subtle noise reduction tool. Fiddling with the many many command-line parameters is cumbersome (given the massive amount of time the processing of even just a short snippet consumes), a cleanup and preferably a set of usable presets sure would help. Plus, probably no one needs all the additional EQ and stereo-width function etc, there are better and simpler to use tools for that kind of stuff.

But above all, Mr.Dyson: please remove that dreaded fixed overall EQ... as long as you don't, the tool is effectively unusable.
Strictly IHMO, your milage may vary....
PS: And no, I won't communicate with Mr.Dyson, given his arrogant stance and the insults he spilled on me.
This post is meant as general information, mainly for others.

 

[Ron Texas]

I'm not sure what to make of this.

 

[KSTR]

@Ron Texas , in the simplest case, just listen to the examples and compare that to the example comparisons provided by Mr.Dyson -- the latter being much harder to do because of the gross EQ he applies which dominates the listening experience.

 

[John Dyson]

I found some time to play with the latest revision, which is V2.2.3E btw, as contained in the archive linked above.

I used the Anne Murray original CD track snippet "Danny's Song" as provided (because it is an excellent and clean production) and applied the decoder, using a batch file:

da-win.exe --info=11 --fz=opt --fw=classical --fa --input="%1" --outgain=0 --floatout --overwrite --output="%2"

After that, I downsampled it from 88.2kHz back to the original 44.1kHz with Adobe Audition to make things easier (could have used SoX as well but all this was done on a Windows machine).

Again, and as expected, the result showed Mr.Dyson's hard-coded personal taste EQ, the same curve we've seen before and which is probably present in any of his examples:
View attachment 122626
The frequency response was obtained by placing a (REW-generated) LogSweep at the start of the track which then was converted to an Impulse Response with REW.
An interesting detail is that the phase does not match the magnitude response at the kinks seen at 3kHz and 9Khz and the irregularities at ~1.5kHz, this suggests the linear transfer function of the decoder is not minimum phase (if it were min-phase the wiggles would be fully reflected in the phase response as well, can be checked by letting REW compute a min-phase version).

This impulse response was used (I) to de-embed the EQ function in the decoded track in order to restore its original frequency response for a meaningful listening comparison, and (II) to apply that same exact EQ to the original so that a full-fleshed subtractive analysis is possible with a result that can be actually listened to (as DeltaWave's built-in FR compensation produces unavoidable processing artifacts).

The corrected FR for listening was created by REW, generating filter parameters, which then where converted to an impulse response with RePhase which in turn was used as a convolution kernel in Adobe Audition. This is a minimum-phase correction so the (small) excess phase contribution is preserved. The corrected FR looks like this, with a tolerance band of 1dB (I didn't bother to correct stuff below 40Hz as there is little content anyway).
View attachment 122630

I could have used my double-precision time-domain convolver as well but the ultimate precision and complete absence of artifacts is not needed here. But I used the convolver to obtain the "linear-transfer-function-equalized" files to feed DeltaWave, for a clean residual. I only used the left channel data for this.

The linarity plot obtained from DW shows a 1:1 transfer for the first 10bits (-40dB) and below that the gain decreases, showing a slight expansion. The plot is not really representative for what's going on because, as explained by Mr.Dyson, there is a bit more going on that just a simple single-band downward dynamic expander, but it shows the tendency:View attachment 122631

More revealing is to actually listen to the residual, where we can find what the decoder "took away" from the original. We can hear that it is mostly recoding noise, reverb tails and general lower level content, whereas the loud voice is about equal in both (and therefore mostly cancels). The residual also has a certain roughness and pumping to it, exposing the dynamic processing.

This also what one can hear (when turning up the volume) in the direct comparison of the original and frequency-corrected "decoded" version. Personally, I'm not sure if I like the processed version, it sounds quite dull in the quiet sections (on the guitar, notably) and all the ambience details that make that (rather clean) recording sound any natural/live are blurred or outright missing, the downward expansion is just too drastic. Stereo image I don't find to be improved, rather the contrary, by this.
The differences are large enough that most people will be able to succesfully ABX this (if anyone feels a need to do so)... just turn up the volume quite a bit.

Link to a ZIP archive containing the following files:
- Original snippet ("raw").
- "Decoded" snippet as per above settings, with restored frequency response as explained.
- Delta (residual file), with +40dB boost to make it audible.

My conclusion. Mr.Dyson's "Decoder" actually does something (besides that nasty gross EQ which basically spoils its usage for non-tinkerers) and it might actually improve some tracks or find its place as a subtle noise reduction tool. Fiddling with the many many command-line parameters is cumbersome (given the massive amount of time the processing of even just a short snippet consumes), a cleanup and preferably a set of usable presets sure would help. Plus, probably no one needs all the additional EQ and stereo-width function etc, there are better and simpler to use tools for that kind of stuff.

But above all, Mr.Dyson: please remove that dreaded fixed overall EQ... as long as you don't, the tool is effectively unusable.
Strictly IHMO, your milage may vary....
PS: And no, I won't communicate with Mr.Dyson, given his arrogant stance and the insults he spilled on me.
This post is meant as general information, mainly for others.

Your numbers are meaningless - obviously don't understand what is going on. it is not about EQ, even though that is a portion of the whole process.

 

[John Dyson]

Here is a simple, real recording, comparison... One (RAW) is the hiss, compression encumbered version -- from a 176k consumer original copy. The other (DEC) sounds more like a studio mix:

https://www.dropbox.com/s/3c6nm2w4a5r4pbc/01-Blue Rondo a la Turk-RAW.flac?dl=0
https://www.dropbox.com/s/luahrp6qyjbyo52/01-Blue Rondo a la Turk-DEC.flac?dl=0

Be careful about info sources, there has been industry push back -- they don't want you to know this. (I was told by industry person to call it my admittedly 'amazing expander' instead of a decoder.) I decided, instead to have integrity and tell the truth. Frankly, the audible expansion effect is generally small. It is all about reorganizing the low levels so that the ambence and other low level aspects are corrected. Sometimes you might see a 1-6dB increase in dynamics (that is, decreased loudness for given peak level), but compression isn't necessarily the only purpose to the fiasco started in the 1980s *I think the 1980s*, could have been before.

If you look at the energy before and after, it is almost the same, except for some expansion effects.
AVERAGE SPECTRUMS PRODUCE NONSENSE COMPARISONS -- but can be used as a guideline if you understand the effects. There is NO STATICALLY MEASUREABLE FREQUENCY RESPONSE...

*Input and output spectrums are meaningless for those who don't know what is going on.* You can give a monkey a spanner to use -- but the use won't be very helpful.

I also just got some encouragement and help from a constructive individual -- FINALLY, the LF EQ has been fixed. There are two ways that they did the EQ in the encoding process -- one was before the encoding, and the other was during the decoding (in-between steps.) I had mistakenly decided to use the all-at-once approach like for the HF bands, but that was wrong. I should have used the EQ per step like the MF bands.

About the EQ -- it is about compensating for the 10dB threshold difference in the midrange, the slow HF boost between 3k and 9k, and another 5dB boost between 9k and 20+k. If they don't pre-compensate, then the results are 'bad' to say the least. The EQ that the decoder does undoes the pre-EQ to keep the compression from being insane. It isn't like the thresholds can easily be changed on the compressor that they used.

 

[John Dyson]

2.2.4D is available.

These fixes were encouraged by two of the project's approx 20 users/supporters, and somehow the LF fix happened when I was discussing the matter with one of these users. I don't know why I couldn't have done the LF EQ correctly before. There has been a LOT of good people helping the project, including the DA project partner. I have added a FEW IDs to the 'using' document, and as I can remember/search through mostly private discussions, will add more.

https://www.dropbox.com/sh/1srzzih0qoi1k4l/AAAMNIQ47AzBe1TubxJutJADa?dl=0

It fixes the minor bass mismatch (we are far into the sub-dB error there now.) I might have hesitated too much about rolloff around 20Hz, but decided to be a bit conservative touching the theoretical' EQ. There is still more than enough EQ to handle suppress less than 10Hz type stuff. (The decoder creates NO new <10Hz stuff itself, and in fact produces very little excess sideband energy from gain control at all, even under cases of extreme dynamics. One reason why I think much about <10Hz is for vinyl, which can and IS FA encoded at times. )
Also, the previous HF EQ correction was *slightly* wrong and created about a -1.5dB error at 20kHz, but more important is the corrected 0.5-1.0dB error at 12->15kHz.

Haven't updated the 'StartUsing' document as the only change is a description of the corrections.

About usage:
There is no need for the '--fa=L' submode, simply because the LF bug that forced using it is now corrected. The LF band (<80Hz) is now under better gain control because of more correct pre and de-emphasis -- so, the cases where there is too much bass are now handled by the internals of the decoder. It wouldn't EVER have been a problem if my reverse engineering on the bass (now, a very simple 500Hz and 75Hz between layers) was correct to begin with. Before, I used a large sequence of EQ after decoding, which became very complex because of the numerous expansion layers with the encoded compensatory EQ for the DolbyA devices. My previous LF EQ was NOT mirror image from the encoding, now it is an image of the encoding process. (These results 'locked-in' -- any error in the intra-layer LF EQ essentially produces unlistenable sound -- think decoding DBX on un-encoded material -- errors can make it THAT bad.)

On normal consumer recordings, I don't think that there is ANY need to vary from the default --coff=-2 and the --fa commands anymore. Since --coff=-2 is the default, you won't usually need to specify it. If the recording has been normalized, or EQed after encoding, then all bets are off. So, we have the EQ and calibration levels now mostly hidden from the user -- HORRAY!!!

This release is on the knife-edge of being needed -- this is a correction of SMALL errors -- much more small than the errors in most recordings anyway.

https://www.dropbox.com/sh/1srzzih0qoi1k4l/AAAMNIQ47AzBe1TubxJutJADa?dl=0

Here is an especially clarifying example -- the original 176k consumer obtained recording, and the decoded version (sounds more like when in the studio.) The major flaw in the decoded version is a slight amount of hiss modulation. However, the cymbals are pretty darned good -- that takes good tracking of the gain control.

RAW:
https://www.dropbox.com/s/3c6nm2w4a5r4pbc/01-Blue Rondo a la Turk-RAW.flac?dl=0

DECODED:
https://www.dropbox.com/s/luahrp6qyjbyo52/01-Blue Rondo a la Turk-DEC.flac?dl=0

 

[John Dyson]

Before making judgments on the project -- one should understand what it is

1) It does NOT master a recording.
2) It undoes a 'homogenization' phase of processing - so different recordings DO sound more different than each other
3) After decoding, the recording will sometimes need a small amount of EQ. The 'homogenization' above is depended upon a a skip-step for mastering.
4) It does LOTS of noise reduction.
5) The amount of gain control can be statically 70dB or more, but dynamically seldom changes more than about 20dB
6) The decoder is NOT statically flat. However, the end results are more accurate than the consumer materials that are decoded.
7) The best way to describe the decoders' actions: unscrambles the dynamics of recordings.
8) Is the software 'precise': YES. In fact, the recent LF correction mitigated gating -- with only a partial dB change in the middle/lows (approx 100Hz.) A certain piano piece manifested gating because of the small LF correction error -- the decoder is exquisitely precise.

Is the decoder an 'expander'? Technically yes..
Does the decoder render a recording CLOSER to a studio mix? YES.
Does the decoder render a recording to VERY CLOSE to a studio mix? SOMETIMES, only sometimes.

IMPORTANT: one of the goals of the decoder is that the 'decoding' be minimally audible. There is a LOT of gain control going on (depending on recordings -- sometimes 100-200 back &forth or more dB/second on the 3kHz and 9kHz bands). A design goal is that there be NO AUDIBLE EXPANSION/COMPRESSION EFFECTS. Per some evidence in other postings, there are claims of no dynamics processing -- therefore the goal is met, even though the observer's conclusion is incorrect.

Here is an egregiously scrambled example, and the rather good unscramble, considering: (Dropbox player sucks)
RAW:
https://www.dropbox.com/s/4889seke7hjnpxs/02 - Take A Chance On Me-RAW.flac?dl=0
DECODED:
https://www.dropbox.com/s/znjerng1aq0ouu6/02 - Take A Chance On Me-DEC-V2.2.4D.flac?dl=0

Might you want to do some EQ? Maybe or maybe not -- but what you would hear when listening to the DECODED version is what was applied to the encoding device. You might actually wish to do some mastering yourself, because the decoder does NOT automatically master anything.

The decoder has a whole suite of 2nd order, 1st order EQ, 1band, 2band, 3band simple (but good) compressors, a limiter (doesn't work well), and dynamic anti-sibilance processors (which do work VERY well.) Do I suggest using these tools? NO -- use a DAW!!! The tools, except for EQ are only for expediency, and the EQ might be helpful for material that had be modified before encoding.

New snippets -- some limited +-1.5dB or so mastering/EQ might be beneficial, but as far as anyone can tell, the results are as good as the source.
(That is, too much or too little bass/treble will be mirrored fairly accurately):

https://www.dropbox.com/sh/tepjnd01xawzscv/AAB08KiAo8IRtYiUXSHRwLMla?dl=0

 

[KSTR]

Nothing new under the sun ;-)

Those who want to effectively use the software should apply post-EQ to get rid of Mr.Dyson's personal and rather drastic "house EQ".

Find attached a convolution kernel (32bit float mono .wav 44.1kHz) which can be used to undo the damage for the latest versions (double-checked with V2.2.4E).

At first, you must downsample the "decoded"track to 44.1kHz. The gain of kernel is set rather low so you have headroom to gain-match to the original file later (you can use DeltaWave to do this automatically for you).
Everything is easily done in Adobe Audition 3.0 (which does also run under WINE when some MS font -- micross.ttf -- is correctly installed at ~/.local/share/fonts).
Set Audition to 32bit processing. Load the correction kernel. Open convolver settings, click Delete, set scale to 1:1, click AddSelection, then click Close (don't click OK). Open processed file, downsample to 44.1 32bit float (F11), then apply convolution (this time click OK only). Match gains manually using RMS statistics, or let DW do it for you.
Alternatively, you can create you own correction kernel, it is quite easy to do with REW and RePhase, and make versions for 88.2kHz and 44.1kHz.

I did a few tests with the new snippets provided as well processing some own material, all are restored to their original frequency response. Then, the subtle effect of the decoder can again be heard quite easily, as reduced low level noise floor on quieter sections (taking away too much detail/ambience IMHO). Some files may have increased stereo width but that is completely unrelated to the Dolby-A "decoder" thing.

 

[John Dyson]

View attachment 122818
Nothing new under the sun ;-)

Those who want to effectively use the software should apply post-EQ to get rid of Mr.Dyson's personal and rather drastic "house EQ".

Find attached a convolution kernel (32bit float mono .wav 44.1kHz) which can be used to undo the damage for the latest versions (double-checked with V2.2.4E).
View attachment 122819
At first, you must downsample the "decoded"track to 44.1kHz. The gain of kernel is set rather low so you have headroom to gain-match to the original file later (you can use DeltaWave to do this automatically for you).
Everything is easily done in Adobe Audition 3.0 (which does also run under WINE when some MS font -- micross.ttf -- is correctly installed at ~/.local/share/fonts).
Set Audition to 32bit processing. Load the correction kernel. Open convolver settings, click Delete, set scale to 1:1, click AddSelection, then click Close (don't click OK). Open processed file, downsample to 44.1 32bit float (F11), then apply convolution (this time click OK only). Match gains manually using RMS statistics, or let DW do it for you.
Alternatively, you can create you own correction kernel, it is quite easy to do with REW and RePhase, and make versions for 88.2kHz and 44.1kHz.

I did a few tests with the new snippets provided as well processing some own material, all are restored to their original frequency response. Then, the subtle effect of the decoder can again be heard quite easily, as reduced low level noise floor on quieter sections (taking away too much detail/ambience IMHO). Some files may have increased stereo width but that is completely unrelated to the Dolby-A "decoder" thing.

All I can say is -- for those who KNOW what studio sound is like -- listen to the examples, esp the Brubeck example, since it is true, natural sound.

As I have noted before, and I guess I'll have to explain loudly:

YOU HAVE DONE NO TEMPORAL ANALYSIS OR SHOWN WHAT IS HAPPENING IN THAT REGARD.
FEW CHANGES ARE TO BE EXPECTED -- because the decoder is working nearly perfectly. Too bad that you arent' measuring anything useful.

You are showing the analysis that I'd expect from someone into the first year of college or maybe a high schooler, maybe not even graduate.
There are more dimensions than your 'long term average'.

Sorry, you speak nonsense. Simply do not understand what is going on.


Basically, your information is nonsense.

OTOH - I am getting requests for more information....

 

[John Dyson]

I was just thinking about some usage information about the decoder. When you use the decoder, some recordings are reverted to an almost unmastered state, or perhaps tweaked to sound good before the FA encoding process. Given that fact, sometimes a bit of tweaking of the output is in order. I will not do that, because as a matter of integrity, the decoder should provide a correct copy of the recording.

(Most of these facts are already known by the current user base)

So, there are a few options to deal with the raw state of the decoded output, especially the usual slight upward tilt in the frequency response. This tilt appears to be in the 9kHz range, so when I listen, I do a small -0.75dB cut at 9kHz. Ideally, this cut should be single pole, but some audio software (e.g. SoX) doesn't conveniently do single pole. If you want to roughly approximate the effects of single pole, you might want to try 2nd order, Q=0.50. That is still more sharp than single pole, but it helps better than not using EQ at all. Sometimes, two EQ are helpful... One 2nd order at 9kHz and one at 12kHz. This might be needed more when using 2nd order EQ instead of 1st order.

I know that the demos might not sound 'finished' -- well they are NOT finished. The 'great homogenizer' (the compression/FA encoding) tends to cover up differences between recordings. Therefore, there will be stronger differences in the sound from recording to recording.

This difference isn't profound, but if you want the best experience, then some times personal mastering is appropriate.

I can say the same thing about bass. The dynamics are correct, but there might be a need to increase or decrease the bass a little -- perhaps by 0.75 or 1.5dB at 250Hz (LF shelf.) When you do a 1.5dB LF shelf at 250Hz, the difference is MUCH more small at 250Hz, and only approaches 1.5dB below 100Hz. BE GENTLE WITH EQ -- esp 1st order EQ. Unless experienced, 2nd order EQ might be easier.

 

[John Dyson]

Future for the FA decoder project.
Already, the decoder trims off some of the modulation distortion from old DolbyA units encode/decodes (called DolbyA fog), but I have come up with a new scheme that might do even better, but not sure. The current scheme is very advanced and use nonlinear operations on analytical functions... The new scheme is simpler, and might not trim off previous modulation distortions, so might be a tradeoff. The new method will be about 2X faster, and mitigate a large part of the modulation distortion locally generated (much less anyway than if a DolbyA unit was doing the processing -- even without the anti-MD code.)

I am also thinking about a super mode, where both the trimming and the new scheme is used. There is the possibility that the combo, with a few design adjustments will be better yet and just as quick (or slow) as the current full quality ode.

One reason why I mention the anti-MD mechanisms (never been done before) is that FA recordings upon decoding will appear more bright than the normal versions. This also happens in DolbyA mode where I do have before and after examples - the DHNRDS sounds very close to the originals, but the DolbyA DOES NOT. The different between DolbyA and DHNRDS DA is profound. The FA mode is like this times 7. In some cases, you cannot even get a real cymbal sound out of a DolbyA, but you can EASILY with the DHNRDS.

Sometimes recordings were done to push a little more clarity through the old DolbyA units. Why do DolbyA units have a 'soft' sound? Two reasons... modulation distortion and a delay in the audio feedback loop when decoding. Encoding is *perfect*, but decoding is not very perfect. This gives the DHNRDS an opportunity to improve over the original - because it has no such delays and DOES trim the modulation effects. (How can that be done, as modulation must happen? Answer, there is a dynamic math operation between the signal and gain control -- thereby avoiding the highest strength sideband generation, deferring it for when the signal is closer to zero crossing -- well not quite, but conceptually similar.)

There are differences, but once getting used to the more original sound -- other consumer recordings sometimes just sound bad. When I say 'getting used to' the FA decoder sound, I mean that the listeners hearing loses the accomodation associated with current normal recordings.

 

[KSTR]

I had been warned before but thought I'd give it try anyway... only to find confirmed what we all aready know: Any discussion with flat-earthers, creationists etc and notably with people encapsulated in their own micro-miniature Dyson Sphere is utterly futile, no matter if they are laymen or experts in one field or another.

The data I presented is real and correct and can be asserted by anyone, all you need is listen to the examples to find out the grossly skewed frequency response. And the technical analysis is correct as well and Mr.Dyson fully knows that, you're not dealing with morons here, Mr.Dyson (and you know nothing about my scientific/engineering/recording/musician etc background). I never claimed the decoder is not "working perfectly", btw.

If the decoder leaves the higher signal levels alone (which it does) then any method of determining its frequency response that uses high enough (> -10dB) signal levels for most of the time and spectrum fully represents what's going on, I've used the music itself, high level sweeps, high level noise. all with the exact same result of that constant +-6dB peak/valley EQ curve (plus those strange kinks) impressed on each and every decoding Mr.Dyson presented and on any decoding I tried myself. Does that run under "few changes"?

Here at ASR any bold claims are put to test and we all agree to the proven fact that large frequency response abberations are the most dominant thing wrt to perception.

I could actually show (and have done already, in the subtractive analysis listening example presented) what the decoder does dynamically to small signals in all details (wtr to frequency band, time behavior etc etc) but who am I to help out Mr.Dyson with data he should have aquired and presented in the first place?
Could it be he never technically analysed what the code is actually doing -- other than using is ears -- , perhaps even lacking the skills to do so? How come the gross EQ thing went unnoticed? Where is the claimed meticulous attention to details with that overall EQ and notably those 3kHz and 9kHz kinks completely ignored?

Rather than replying in any meaningful matter to the observations made (not only by me, btw) the only answer Mr.Dyson is capable of is repeated insults and ad-hominem attacks... strong enough to get called out by a mod if I just cared enough...

 

[mansr]

I had been warned before but thought I'd give it try anyway... only to find confirmed what we all aready know: Any discussion with flat-earthers, creationists etc and notably with people encapsulated in their own micro-miniature Dyson Sphere is utterly futile, no matter if they are laymen or experts in one field or another.

I have come to the conclusion that this project started out with good intentions but soon turned into a wild goose chase down a rabbit hole. There may well exist releases with unintentional Dolby encoding. I do, however, doubt the problem is quite as prevalent as Mr Dyson would have us believe. The meandering development of this software has all the signs of someone attempting to fix a problem that is either non-existent or misunderstood, tweaks piled upon tweaks until all the test cases pass for reasons unknown. I suspect that by now the software is actually little more than an obfuscated database of Mr Dyson's favourite EQ curve for every track in his collection.

 

[John Dyson]

I have come to the conclusion that this project started out with good intentions but soon turned into a wild goose chase down a rabbit hole. There may well exist releases with unintentional Dolby encoding. I do, however, doubt the problem is quite as prevalent as Mr Dyson would have us believe. The meandering development of this software has all the signs of someone attempting to fix a problem that is either non-existent or misunderstood, tweaks piled upon tweaks until all the test cases pass for reasons unknown. I suspect that by now the software is actually little more than an obfuscated database of Mr Dyson's favourite EQ curve for every track in his collection.

Would you like to see some source code snippets? No database..
Sometimes messy, reworked code -- yes... No database at all. It works with almost all recordings -- no tweaks.
This claim makes me worry about YOUR OWN credibility..


Have you tried to listen to recent results?
Getting really good reviews from unbiased people!!!

Take a look at a spectogram -- anyone with ANY competency can see that there is dynamics processing -- answering silly claims about just being EQ.. In fact, one recent fix was erroneous EQ -- the decoder, if there are internal problems, can easily create gating. The recent gating problem (not seen by users) came from the new bass eq with one of the two parameters set incorrectly. (Usually, professional EQ freqs in the FA system are like 500Hz, 250Hz... on down to 25Hz.) I was confused between the DolbyA 80Hz and the FA standard 75Hz. THAT is the kind of tweaking involved, not like YOUR kind of tweaking where itmight be 81Hz, or maybe even 117Hz... Got it -- that just DOES NOT or HAS NOT happened. Everything are the even numbers (in the sense above) that an engineer would typically use. The even frequency thing is about distortion suppression and things like that -- if you dont use even numbers -- then you can end up with a worse mess of distortion.

The decoder is probably more complex and does more things CORRECTLY than you can ponder. In fact, uses the same tech as the DolbyA decoder, same developer, and it took me a LONG TIME to do that also -- BECAUSE OF REVERSE ENGINEERING. Some people might like the DolbyA mode because it is more clean, otherwise some people might have troubles with older recordings because they were done with the assumption of DolbyA HW fog to begin with.

NOTE: one of the bits of evidence about the decoder being 'true' -- the parameters are quite critical, and the audible effects are often horrendous (like sounding even worse than FA, instead of the clean-up improvement of the decoder.) In fact, without the BASS EQ, the result almost sounds as bad as DBX trying to decode unencoded material. The BASS EQ has to be VERY precise, or the result is -- just TERRIBLE -- ALL BASS or NO bass. What is the bass EQ the fits the key in the keyhole? 500Hz -3dB, 75Hz -3dB (or is it -6dB -- I forget) on each layer. If you use 400Hz or even 520Hz, the result is really bad.

Key in a keyhole, hand-in-hand -- that is how critical the iinternal settings are. These setting are ALL EVEN NUMBERS. NO TWEAKS, NO FRACTIONAL VALUES. I cannot believe that anyone with any engineering competency cannot see that the sensitivity to precision isn't clear evidence (but admittedly not proof.) Unlike some people, I DO have integrity (both competency and honesty.)

There are very few tractional/traditional tweaks, except in the DolbyA decoder section, which is emulating selected semiconductor hardware (including JFETS -- which have wide parameter differences.)
The big time consuming, and patience testing problem has been 'reading the mind' of the original developer. (figurative)

Apparently FA was not intended to be decoded, but instead a stamp on the recording to maintain IP control of the original. Do you think that the owners really want to give you effective 'originals'? Heh...

A true DolbyA would probably create a mess because of modulation components (I hope you know what those are -- any 1st or 2nd year EE knows why that happens.) The DHNRDS has a mechanism to suppress all but approx 16% of them. (Higher quality modes can do twice as good.)

There HAS been industry push back because of the IP issues. Also, the decoder can overcome the lost 2 bits from MQA. This is because those bits are effectively noise, and it is well below the point where it would be the hiss that is so beloved by legacy stereo lovers.

The last problem -- the bass, was fixed by a much simpler design.
Some people are comforted by hiss and only 0deg and 90deg stereo images. (FA encoding tends to suppress 45deg.)

 

[mansr]

I understand you think very highly of yourself. Unfortunately, arrogance offers no protection against being wrong, only against admitting it.