'FeralA' decoder -- free-to-use

[John Dyson]

56 minutes ago, PeterSt said:

 

I am totally clueless as to what you want to tell me there.

Maybe someone can translate it ?

 

Sorry -- my stream of consiousness has nothing to do with English.   Let me explain what I was trying to say.

 

If you would have given me whatever full information that you had, with some kind suggestions (instead of an apparent set of evidence against), then maybe I would have made progress more quickly.  It was a kind person who gave me enough information that I found that my perception of HF was wrong, so I stopped my very imporant medicine so that my hearing would come closer to normal.   Thank you for your lack of help, thereby necessitating a risk to health.

 

My hearing of envelope intermodulation is very well trained by my past and my development of the only plausible SW DolbyA decoder.   That decoder has explicit mechanisms to avoid generating more of the psuedo-IMD that I speak of in the next statements.

 

For those nay-sayers -- listen carefully to Nat King Coles vocals.  i don't have to listen carefully, but probably most people do have to listen carefully.

Listen carefully to a kind of inter-modulation of the envelope of his vocals on the FA material.   That modulation gives an impression of 'detail', but just like hiss, it isn't more detail.

 

Listen to the decoded version -- notice the lack of that intermodulation of the envelope... (please ignore the slightly heavy lower midrange -- I am looking to follow the rules and improve the situatiion.

 

This pseudo-IMD comes from the various envelopes interacting - caused by the super fast compression -- made worse because of MULTIPLE layers of compressioin, and not a natural part of the voice.

 

This pseudo-IMD is one of the components of 'garble' that I sometimes mention.

 

 

 

 

 

[PeterSt]

5 minutes ago, John Dyson said:

Thank you for your lack of help, thereby necessitating a risk to health.

 

WTF ?

 

I need a way better (off) translator to understand this in a positive sense.

[MarcelNL]

I'd really appreciate if the discussion would remain on the topic and not deviate to 'on the person'.

[John Dyson]

Just now, PeterSt said:

 

WTF ?

 

I need a way better (off) translator to understand this in a positive sense.

You are very quick to criticize, arent you..

I think that you can parse my statement...

 

BTW -- read the substantive statement that I made about pseudo-IMD (that is, the compression gains are flopping around, based on signal envelopes, thereby modifying the gain of both parts associated with the signal, and other parts of the signal also.)   This is much less of a problem on normal compressors -- yet more evidence of likely DolbyA involvement.

 

[John Dyson]

1 minute ago, MarcelNL said:

I'd really appreciate if the discussion would remain on the topic and not deviate to 'on the person'.

Yes - I will -- just look at my statement about pseudo-IMD (envelope intermodulation based on gain control)  that it is evidence of a super fast compressor, not a guarantee of DolbyA though.

 

[PeterSt]

2 minutes ago, John Dyson said:

You are very quick to criticize, arent you..

 

Yes, you are off track. And hopelessly lost with it.

[John Dyson]

5 minutes ago, PeterSt said:

 

Yes, you are off track. And hopelessly lost with it.

I suggest that you look at the substantive comments about gain control modulation of vocals and other things.   This is a problem only on very fast compressors that work at envelope rate speeds.   A normal 250msec release compressor or slower won't really do that very much.   Therefore, evidence of DolbyA like attack/release technology in the FA signal.   It isn't proof, but is pretty strong evidence of DolbyA.  This is because of the timeframe that FA was started, I doubt that there were many compressors that used the 'tricks' to the extent of a DolbyA HW unit.

 

 

[Jud]

16 minutes ago, John Dyson said:

Thank you for your lack of help, thereby necessitating a risk to health.

 

That kind of thing is just "not on," to use a British-ism. Responsibility for your decisions is yours, John.

 

Agreed with MarceINL that we should be talking audio projects here, which to me is the interesting stuff.

[John Dyson]

8 minutes ago, Jud said:

 

That kind of thing is just "not on," to use a British-ism. Responsibility for your decisions is yours, John.

 

Agreed with MarceINL that we should be talking audio projects here, which to me is the interesting stuff.

I was only frustrated with information not provided by those who criticize. 

I am trying to get back on target.  I truly suggest reading about the 'gain control modulation' of the signal.   The gain control 'wobbles' around so much that other parts of the signal get modulated.   It is a kind of 'buzz' around various parts of the signal.   When that buzz adds up too much, the you get the kind of sound on the 'Take a chance on me' recording.

 

The decoder removes the buzz by counteracting the gain control in the encoding process.  If the counter-acting wasn't very accurate, then the buzz could get worse.  I think that 'take a chance on me' is at least a little better after decoding.   The accurate DolbyA decoding is the only way that can happen.  (Even a normal DolbyA might have problems because of DolbyA fog -- yet another interaction with gain control.)   DolbyA fog is one reason why people sometimes (really only sometimes) use the decoder, because it has almost NO fog.

 

 

 

[PeterSt]

5 minutes ago, John Dyson said:

I was only frustrated with information not provided by those who criticize.

 

As you did not say Thank You Peter, I think you are making a fool of yourself.

Let's take a break and let your meds work-in or work-out.

[John Dyson]

Just now, PeterSt said:

 

As you did not say Thank You Peter, I think you are making a fool of yourself.

Let's take a break and let your meds work-in or work-out.

My BP is coming down - btw - did you look at the technical issues?  Made up your mind?

 

[PeterSt]

13 minutes ago, John Dyson said:

btw - did you look at the technical issues?  Made up your mind?

 

Yes. My IQ is too low.

I your AHI all right now ?

[John Dyson]

Oh -- about 'take a chance on me'   Because of speed issues, I used the lowest 'high quality' setting.   I tried the highest normal 'high quality' setting, and it sounds better yet.

I'll attach a snippet of it... Sorry about the wierd filename -- I did this quickly.   It IS a flac file.  I'll update the demos in about 10minutes.

 

The previous demo was 'better', but I believe 'no cigar'.   This one uses more anti-garble technology, but also takes more CPU.   It is incredibly CPU intensive, and

I really need to find a math expert who can help to simplify the algorithms.   I could give a conceptual implementaion, but like a DFT vs FFT, there is probably a better way of doing the anti-garble, anti-fog, etc. (still runs realtime, but just barely -- that is one huge amount of CPU.)

 

One more thing -- some versions of the demos have 'replay gain' enabled.  I am uploading (again) to disable it.

 

 

 

02 - Take A Chance On Me-V2.2.5F-0.wav.flac

[John Dyson]

33 minutes ago, John Dyson said:

Oh -- about 'take a chance on me'   Because of speed issues, I used the lowest 'high quality' setting.   I tried the highest normal 'high quality' setting, and it sounds better yet.

I'll attach a snippet of it... Sorry about the wierd filename -- I did this quickly.   It IS a flac file.  I'll update the demos in about 10minutes.

 

The previous demo was 'better', but I believe 'no cigar'.   This one uses more anti-garble technology, but also takes more CPU.   It is incredibly CPU intensive, and

I really need to find a math expert who can help to simplify the algorithms.   I could give a conceptual implementaion, but like a DFT vs FFT, there is probably a better way of doing the anti-garble, anti-fog, etc. (still runs realtime, but just barely -- that is one huge amount of CPU.)

 

One more thing -- some versions of the demos have 'replay gain' enabled.  I am uploading (again) to disable it.

 

 

 

02 - Take A Chance On Me-V2.2.5F-0.wav.flac 4.92 MB · 1 download

WOW!!  I found the lower MF problem (about 100-200Hz!!!)   That was the last major EQ problem.   This was  a weird fix, but still always 'follow the rules'!!!

there will be an V2.2.5G soon.

Thank gosh!!!!

 

[John Dyson]

GOOD NEWS -- V2.2.5G demos are up on the site

MIDRANGE MUCH BETTER -- tell me how it needs to be moved around.  I currently moved the EQ from 75Hz

to 50Hz and 37.5Hz (Yes, standard values).   If 75HzEQ slops over past 100Hz, the sound gets overly heavy.

 

I have at least a few warnings about differences to expect:

1) The bass is tighter.  That is just an artifact of expansion, and I fully expect that the RAW FA sloppy bass is because of compression

2) Vocals are sometimes very forward, sometimes a little forward, sometimes about the same as the RAW versions.  It is tricky to predict what the expansion will do, because the decoding is in M+S or M+2S.   The forwardness (:-)) has little to do with EQ, moreso caused by expansion (the reason for the variability)

3) Vocals have less of the nasal, telephone type quality -- you might like the improvement, you might not.  It is all about what you like.

4) The recordings have some variations in level.  That is hard to fix, because after expansion, the result can be louder or softer.  All are normalized.

5) Generally expect the transients to be a little stronger, but YMMV.

6) No change in stereo image is intentional -- that is just the way that it decoded. Often, the recordings will become alittle more wide.

 

There are two possible known  issues:

1) I might have still missed the midrange, but I doubt it is very far off (I used standard values)

2) The vocals might still be a little bright, you know that I have a LOT of freedom given the standard values -- I can darken the vocals pretty easily still following the rules.

3) Tell me more.  I will NOT bite your head off, but I really need specifics, not just 'sucks'., or just disagree -- 'sucks' or 'disagree' are not helpful -- explain what bothers you as I can DEFINITELY fix it.

 

Here are the demos: (the direct dropbox player tends to scramble the sound -- if you listen directly, just remember that!!!)

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

 

The decoder willl be coming once we settle this down.

 

 

[John Dyson]

I wanna explain another attribute of 'garble' and it is an important one...

If you listen to 'Take A Chance On Me', the raw version has a chorus, but you cannot pick out the singers at all -- they are all garbled together.

The decoded version does better doing a straight line to at least a group of singers or even one.  However, there are higher modes that can do

even more, but they are painfully slow and reserved for DolbyA.  Since DolbyA uses only one virtual unit, the decoding is much faster, so the better modes can be used.

 

If you really want, and want to take 3Hrs per album, you CAN use the higher modes on FA, myself, I don't have the patience.

The first difference is that the normal FA modes THEORETICALLY slice off about 85% of the gain control IMD, while the 'professional' modes slice off

about 92%.    You can REALLY hear the differences in singing chorus on the DA mode, but the FA mode has natural precision

errors between layers.   Also, there are LOTS more gain control distortion sidebands -- 'FOG', that the subsequent decoders have to deal  with.  Even though I said that the decoder has less fog than a DolbyA, it really doesnt'.  It moves the fog around, and kind of bends the gain control signal around the audio signal.   The sidebands are created at when less audible.   The total energy is the same.  If you try to hard code this without using the analytic version of the signal, all you'll get is signal jitter and distortion.  (A analytic version of the signal is the original real signal along with the Hilbert transform (90deg) version.)

 

So, there are real limitations about the FA cleanup - and I probably wouldn't expect more than 50% to 75% improvement in FA modes.

 

 

 

[KSTR]

OK, I see my previous posts have been restored, at least the textual section. Thanks for that to all who supported this.

 

I've now made a series of measurements to show the level-dependent characteristics of the decoder (V2.2.4E). Test stimulus was (periodic) Pink Noise at various levels, starting at -3dB(rms) down to -73dB(rms) in 10dB steps. Pink noise does resemble typical music signal quite well, notably very close to reverb tails and such. It is static, though, it does not have dynamic changes so temporal effects (attack and decay times of the expander) will not show up -- time domain effect example will be shown in the last paragraph.

 

Command line (in a batch file) to create the decoder outputs:

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

 

 

First, a set of curves that represent the output signal vs input signal over the various levels:

Since we are interested in the ouput vs input characteristic allowing a direct reading, the input has to be flat. This was achieved by using 1/48th octave bins in the spectum instead of the direct FFT bins (which are spaced evenly at x.yHz intervals) so that the pink spectrum (with its -10dB/decade slope) is rendered flat, the corresponding trace is labelled "Input".

 

We can see that the first three levels (at -10dB, -20dB and -30dB relative to the -3dB(rms) input) have the known shape (spanning some 12dB of level change and with those kinks at 3kHz and 9kHz mentioned previously), let's call it the "JD house curve", and they are evenly spaced 10dB apart (the dB levels in the legend are taken at 300Hz and we can see the corresponding -3.9dB, -13.9dB and -23.9dB level point sequence). This means there is basically no action from the decoder (other than the house curve).

 

At -30dB and below, additional effects are introduced: the shape vs. frequency changes slightly, notably at the low bass and high treble, and the curves are now spaced more than 10dB apart. This means now there is downward expansion and it is frequency dependent.

 

 

A better visual representation of that frequency-dependent downward expansion is achieved when we plot the gain reduction vs frequency over the set of input levels:

This plot is "normalized" to the output signal of the -3dB(rms) input signal so that the constant house EQ (as measured at higher levels) is factored out, so we can see the actual gain reduction. In other words, it is normalized to the "IN: 0dB" trace of the previous plot.

The dB values in the legend show the gain reduction at 1kHz.

 

The first two lines, for input levels down 10dB or 20dB repectively, again show (and must show, it's the same data as before) that there is no gain reduction at all -- maybe a hint (0.2dB) of expansion for the dark green -20dB curve below 100Hz.

 

At -30dB, the very low bass is reduced by 4dB and interestingly there is a slight gain increase at 100Hz.

At the subsequent lower input levels, we see more and more gain reduction the lower the signal is (reaching 50dB of gain reduction at 1kHz for a -70dB input) and also that the overal curve shape changes, showing the multi-band nature of the expander with different expansion profiles in the individual bands.

 

This relates very nicely to what happens to a slowly decaying reverb tail, it is dying overproportionally in level vs. the input and even more so at the high treble. Likewise, a constant pink noise floor in the recording of -70dB is reduced by at least 50dB making it totally disappear. This all happens multi-band, that means for example a high signal level in the bass band does not keep the decoder from expanding and reducing noise elsewhere according to the signal levels in these other bands.

 

 

Finally, a short look at the temporal behavior, showing the settling to the new charactistic when dropping a pink noise signal from -23dB(rms, input) down to -63dB in a step change:

Here we can see that it takes roughly 250ms to settle to the new output for this example, basically morphing from the -20dB characteristic to the -60dB characteristic shown above.

 

I hope these visual representations help to better understand the basic function of the decoder. While Mr.Dysons textual explanations are very elaborate and extremly detailed I feel that the overall picture of the decoder's working is easily lost in all that details.

[fas42]

5 hours ago, John Dyson said:

 

For those nay-sayers -- listen carefully to Nat King Coles vocals.  i don't have to listen carefully, but probably most people do have to listen carefully.

Listen carefully to a kind of inter-modulation of the envelope of his vocals on the FA material.   That modulation gives an impression of 'detail', but just like hiss, it isn't more detail.

 

 

John, would you be able to point to a specific, "worst" example of this? You may have a snippet uploaded already, otherwise just a track name, perhaps.

 

Thanks!

[John Dyson]

56 minutes ago, KSTR said:

OK, I see my previous posts have been restored, at least the textual section. Thanks for that to all who supported this.

 

I've now made a series of measurements to show the level-dependent characteristics of the decoder (V2.2.4E). Test stimulus was (periodic) Pink Noise at various levels, starting at -3dB(rms) down to -73dB(rms) in 10dB steps. Pink noise does resemble typical music signal quite well, notably very close to reverb tails and such. It is static, though, it does not have dynamic changes so temporal effects (attack and decay times of the expander) will not show up -- time domain effect example will be shown in the last paragraph.

 

Command line (in a batch file) to create the decoder outputs:

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

 

 

First, a set of curves that represent the output signal vs input signal over the various levels:

Since we are interested in the ouput vs input characteristic allowing a direct reading, the input has to be flat. This was achieved by using 1/48th octave bins in the spectum instead of the direct FFT bins (which are spaced evenly at x.yHz intervals) so that the pink spectrum (with its -10dB/decade slope) is rendered flat, the corresponding trace is labelled "Input".

 

We can see that the first three levels (at -10dB, -20dB and -30dB relative to the -3dB(rms) input) have the known shape (spanning some 12dB of level change and with those kinks at 3kHz and 9kHz mentioned previously), let's call it the "JD house curve", and they are evenly spaced 10dB apart (the dB levels in the legend are taken at 300Hz and we can see the corresponding -3.9dB, -13.9dB and -23.9dB level point sequence). This means there is basically no action from the decoder (other than the house curve).

 

At -30dB and below, additional effects are introduced: the shape vs. frequency changes slightly, notably at the low bass and high treble, and the curves are now spaced more than 10dB apart. This means now there is downward expansion and it is frequency dependent.

 

 

A better visual representation of that frequency-dependent downward expansion is achieved when we plot the gain reduction vs frequency over the set of input levels:

This plot is "normalized" to the output signal of the -3dB(rms) input signal so that the constant house EQ (as measured at higher levels) is factored out, so we can see the actual gain reduction. In other words, it is normalized to the "IN: 0dB" trace of the previous plot.

The dB values in the legend show the gain reduction at 1kHz.

 

The first two lines, for input levels down 10dB or 20dB repectively, again show (and must show, it's the same data as before) that there is no gain reduction at all -- maybe a hint (0.2dB) of expansion for the dark green -20dB curve below 100Hz.

 

At -30dB, the very low bass is reduced by 4dB and interestingly there is a slight gain increase at 100Hz.

At the subsequent lower input levels, we see more and more gain reduction the lower the signal is (reaching 50dB of gain reduction at 1kHz for a -70dB input) and also that the overal curve shape changes, showing the multi-band nature of the expander with different expansion profiles in the individual bands.

 

This relates very nicely to what happens to a slowly decaying reverb tail, it is dying overproportionally in level vs. the input and even more so at the high treble. Likewise, a constant pink noise floor in the recording of -70dB is reduced by at least 50dB making it totally disappear. This all happens multi-band, that means for example a high signal level in the bass band does not keep the decoder from expanding and reducing noise elsewhere according to the signal levels in these other bands.

 

 

Finally, a short look at the temporal behavior, showing the settling to the new charactistic when dropping a pink noise signal from -23dB(rms, input) down to -63dB in a step change:

Here we can see that it takes roughly 250ms to settle to the new output for this example, basically morphing from the -20dB characteristic to the -60dB characteristic shown above.

 

I hope these visual representations help to better understand the basic function of the decoder. While Mr.Dysons textual explanations are very elaborate and extremly detailed I feel that the overall picture of the decoder's working is easily lost in all that details.

actually, I have never seen the diagrams and thank you for them.

 

The newer version is better with the HF/LF bounds -- I traded off some meds for hearing a few days ago, and now understand some of the criticism -- mostly fixing things.

 

The decoder is NOT flat in the normal sense and should never be.   You might not agree, but the consumer audio signal (FA) has already been processed with the complementary compressor.   The decoder attempts (cannot ever be perfect) to recover the original signal.   One (correction: SOME)  of my previous posts tries to give an idea about the challenges.   The gain control pseudo-IMD (it is a kind of envelope interaction) is fairly well compensated for in the decoding process, but unless I get everything right, it can never be perfect.

 

(correction -- comment in the wrong place -- I get brain glitches all of the time -- you wouldn' believe my out-of sync typing!!!_)  Here is an example of the sensitivity of the interaction -- between each unit there is a 500Hz -3dB and 75Hz -3dB 1st pole EQ filter.  If it is off by just a little bit, you loose almost all  bass or get your hearing blown away.

 

Also, there is another kind of thing happening -- when the gain control interacts with the signal, then sidebands are directly created.   Historically, that had been called DolbyA fog.  This happens a lot more with DolbyA because of the extremely fast attack release and the huge amount of gain control happening.  This creates a lot of FOG/sideband energy.  Such 'fog' can create errors in the  multi layer decoding.  The DA/FA decoder DOES create fog, but hides it very well.  Also, here are immediate remediations which keep it from generating as much fog to begin with -- but not as effective as anyone likes withou the extremely CPU intensive algorithms.

 

When there are 7 layers (that is the only number that works), there is a lot of transmission of distortion sidebands transmitted through the system.   The decoder is MUCH better at doing its 'thing' on a DA, but is still partially effective on FA.   I doubt that I true DolbyA HW could do quite as well as the DHNRDS decoder -- even though there are accuracy errors at different  levels and frequencies on the DHNRDS.   it is just that they are mostly not important.  (there is approx 0.35dB flatness error to begin with.)  This is not directly related to the FA mode lack of flatness.  It is an internal DA decoding error.

 

Why are the FA decoder not really flat -- there are a lot of interactions between layers.  For example, the MF band doesn't start doing gain control until at least -10dB below the other bands.   The LF band is about -1.5dB off (AFAIR.)   Also, the HF1 band (9kHz on up) directly interacts with the HF0 band, because the HF1 band cascades onto the HF0 band.   The compression ratio is twice as high at 9kHz->20kHx as 3kHz->9kHz.  There are a hell of a lot of things happening, along with a 10dB difference between layers.  So, if one virtual DA is flat, then with the 10,20,30dB differences, then the other ones will not be.

 

This is a complex thing -- and has taken about 5yrs of reverse engineering, and still probably havent gotten it 100% right.

 

There is a strong reason for my sensitivity, even more than the DolbyA compatible decoder, I got a HUGE amount of industry resistance to the FA decoder.  It was fairly intense, and even though there were admittedly other reasons, probably the biggest reason is my defensiveness is about the industry pushback.

 

Anyway, thanks for the technical information, it was interesting to me.

 

John

 

[John Dyson]

6 minutes ago, fas42 said:

 

John, would you be able to point to a specific, "worst" example of this? You may have a snippet uploaded already, otherwise just a track name, perhaps.

 

Thanks!

Take a chance on me (ABBA).   Since it is so bad, the decoder isn't quite as effective, but still does a fairly good job.  It is like anything else, if it isn't needed, then it does its job better :-).

A hint about detecting what I am talking about -- try to chase the choral groups down (distinguish them) on the RAW version.   On the RAW version, the chorus is almost 'smushed' together.  Then notice that there is more coherency, independence of each singer/group on the decoded version.  Like I said before, since it is so bad, the decoder is less effective.  When there are so  many disturbances, the decoder can do only so much.

 

The main reason why I included it in the examples is exactly this fact.

 

 

[fas42]

Just now, John Dyson said:

Take a chance on me (ABBA).   Since it is so bad, the decoder isn't quite as effective, but still does a fairly good job.  It is like anything else, if it isn't needed, then it does its job better :-).

A hint about detecting what I am talking about -- try to chase the choral groups down (distinguish them) on the RAW version.   Then notice that there is more coherency, independence of each singer/group on the decoded version.  Like I said before, since it is so bad, the decoder is less effective.  When there are so  many disturbances, the decoder can do only so much.

 

The main reason why I included it in the examples is exactly this fact.

 

 

 

Sorry, John, I should have worded my request more carefully ... I was after a Nat King Cole clip, that shows the behaviour you talk of in my quote.

[John Dyson]

16 minutes ago, fas42 said:

 

Sorry, John, I should have worded my request more carefully ... I was after a Nat King Cole clip, that shows the behaviour you talk of in my quote.

I heard it especially strong on one of the 3?.   I promise tomorrow I'll look it up and post it.   I am nearing the end of my day, and have had a hard week.  However, I will definitely try to show how there is a bit of false edginess on the vocals.  it might really be hard to hear at first...   Even though I have problems with HF/LF, a lot of the energy is in the region where I hear well.  Also, I had to train my hearing for exactly that kind of distortion (recording back when younger and doing the DA/FA decoders.)   But, I will remember this, and get back with you.  It might be 24Hrs because of a Dr appt.

Thanks for asking.

 

ADD-ON:   I must give a caveat -- since the lower MF is better(less), the edginess might be a bit more difficult to hear.

 

[John Dyson]

1 hour ago, KSTR said:

OK, I see my previous posts have been restored, at least the textual section. Thanks for that to all who supported this.

 

I've now made a series of measurements to show the level-dependent characteristics of the decoder (V2.2.4E). Test stimulus was (periodic) Pink Noise at various levels, starting at -3dB(rms) down to -73dB(rms) in 10dB steps. Pink noise does resemble typical music signal quite well, notably very close to reverb tails and such. It is static, though, it does not have dynamic changes so temporal effects (attack and decay times of the expander) will not show up -- time domain effect example will be shown in the last paragraph.

 

Command line (in a batch file) to create the decoder outputs:

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

 

 

First, a set of curves that represent the output signal vs input signal over the various levels:

Since we are interested in the ouput vs input characteristic allowing a direct reading, the input has to be flat. This was achieved by using 1/48th octave bins in the spectum instead of the direct FFT bins (which are spaced evenly at x.yHz intervals) so that the pink spectrum (with its -10dB/decade slope) is rendered flat, the corresponding trace is labelled "Input".

 

We can see that the first three levels (at -10dB, -20dB and -30dB relative to the -3dB(rms) input) have the known shape (spanning some 12dB of level change and with those kinks at 3kHz and 9kHz mentioned previously), let's call it the "JD house curve", and they are evenly spaced 10dB apart (the dB levels in the legend are taken at 300Hz and we can see the corresponding -3.9dB, -13.9dB and -23.9dB level point sequence). This means there is basically no action from the decoder (other than the house curve).

 

At -30dB and below, additional effects are introduced: the shape vs. frequency changes slightly, notably at the low bass and high treble, and the curves are now spaced more than 10dB apart. This means now there is downward expansion and it is frequency dependent.

 

 

A better visual representation of that frequency-dependent downward expansion is achieved when we plot the gain reduction vs frequency over the set of input levels:

This plot is "normalized" to the output signal of the -3dB(rms) input signal so that the constant house EQ (as measured at higher levels) is factored out, so we can see the actual gain reduction. In other words, it is normalized to the "IN: 0dB" trace of the previous plot.

The dB values in the legend show the gain reduction at 1kHz.

 

The first two lines, for input levels down 10dB or 20dB repectively, again show (and must show, it's the same data as before) that there is no gain reduction at all -- maybe a hint (0.2dB) of expansion for the dark green -20dB curve below 100Hz.

 

At -30dB, the very low bass is reduced by 4dB and interestingly there is a slight gain increase at 100Hz.

At the subsequent lower input levels, we see more and more gain reduction the lower the signal is (reaching 50dB of gain reduction at 1kHz for a -70dB input) and also that the overal curve shape changes, showing the multi-band nature of the expander with different expansion profiles in the individual bands.

 

This relates very nicely to what happens to a slowly decaying reverb tail, it is dying overproportionally in level vs. the input and even more so at the high treble. Likewise, a constant pink noise floor in the recording of -70dB is reduced by at least 50dB making it totally disappear. This all happens multi-band, that means for example a high signal level in the bass band does not keep the decoder from expanding and reducing noise elsewhere according to the signal levels in these other bands.

 

 

Finally, a short look at the temporal behavior, showing the settling to the new charactistic when dropping a pink noise signal from -23dB(rms, input) down to -63dB in a step change:

Here we can see that it takes roughly 250ms to settle to the new output for this example, basically morphing from the -20dB characteristic to the -60dB characteristic shown above.

 

I hope these visual representations help to better understand the basic function of the decoder. While Mr.Dysons textual explanations are very elaborate and extremly detailed I feel that the overall picture of the decoder's working is easily lost in all that details.

Another reason why I don't want to proclaim that the decoder can ever be normally flat -- lots of EQ is going on.

I truly do not know, but I tend to think dynamically when talking about freq response, so there might be a signal level that SHOULD be nominally flat -- but I am not sure if it really can be.

 

The EQ that I found was based (hopefully) on audible flatness.   This is because with all of the EQ, I don't know how flat it should be.

I KNOW how flat and the frequency characteristics of DolbyA, because I have a schematic (actually three, where one is very dissimilar) and lots of test materials.

 

On FA, I have NOTHING other than engineering guesses and trying to get into the head of the designer.

 

OTOH -- the decoder sounds relatively the same on different numbers of layers, but the difference is that it only sounds 'right' at 7.

So, the EQ doesn't appear to change all that much when the number of serial DA units are used.)   Again, I am working on 'reverse engineering'

because I don't know the criteria at all (I'll bet no-one alive does.)

 

I just wanted to be totally clear about this.

 

[MarcelNL]

on a fairly quick comparison of several snips the difference is quite real to me, the raw sounds more dynamical across f.e. a whole instrument, which in my experience usually means there is room for improvement as there is a 'gray and veiled' character to those events.

Most, and definitely analog, instruments have a leading edge, a ground note(s), overtones and a trailing edge. This is a simplification of course as those sounds will mix in time and space and will interact, and each component has its own time/space trajectory....yet the basic principle is how I envision sound.

When I hear a recording played back where a piano sounds as a single Ploink I know there is work to do, once I get a sense of the attack (can guess at the hardness of the felt tipped hammer striking the snare), a sense of the snare vibrating over it's total length, overtones playing in the air above like butterflies and a decay that is carried by the acoustic of the original room the recording took place I'm getting close.

 

To me what the decoder does is to allow recreating just all of that better, at the trade off of perceived lower dynamics perhaps, but to use a trumpism;  IMO that is 'fake dynamics' caused by sounds squeezed together in time/space for the reason of ......? by the record company, In some cases they might have been clueless, in some they may have attempted to 'spice up the energy' etc.

 

I hate that I'm currently unable to listen in stereo as for me phase is much more important than a ruler flat freq. curve, from what I have heard the freq. curve appears to be close enough (though I'd probably need a couple of sessions with recordings I know better in raw and dec form to be sure) 

 

If the proposition is to have the choice to listen to recordings I'd currently not likely would play because they sound horrible (most of the raw versions sound pretty poor) I prefer the dec in most cases, which would mean albums can be added to the list of options to listen to.

To me that is an option that is a nice to have, yet I'm interested to hear what others think.  

[Don Hills]

9 hours ago, John Dyson said:

Here is what happened -- FA processing isn't really flat like DolbyA is at high levels.   It is weird.  I'll explain FA -- and also recognizing that a low level multi band process will never be flat.   (Esp with the varying thresholds on different bands like on DolbyA, and the units being cascaded at different calibration levels.)

...

Some genius figured out a way to cascade a bunch of DolbyA units to get a very wide dynamic range compressor of some kind.

...
What is going on other than FA existing?

 

 

Thanks, John.

Even assuming that someone did build a compressor using cascaded Dolby A encoders, it can't be in common use. If it was as widely used as you believe, they would be ubiquitous - found in almost every studio. A secret that big could not be kept. And because they would be analogue and work in real time, they would be time-expensive to use, especially when working from digital masters.

 

Knowing what you do about the processes involved, how would you go about building / coding such a FA encoder?

 

Returning to the non-flat response at high levels, can you clarify for me? Is it due to EQ applied at each stage so that the processing works correctly, or is it an overall static EQ? If static, can it be adjusted flat using command line parameters, or would I need to apply external EQ?

 

We really need more examples for comparison - masters before FA processing versus the released CDs. I understand why they are hard to come by.