John Dyson

So far, good news, got a few responses -- not all totally positive. There are apparently two kinds of problems -- the HF detail is a bit amorphous, like a blob and a bit too strong... This is where the slightly strong 'lower HF' comment came from. Also, the bass hasn't matched the FA version very well. I have been doing research on both issues, and the decoder can easily adapt to fix the concerns. However, having the capability, and finding the most accurate solutions are two different things. The new discriminator gives oodles of freedom, and can easily overshoot the HF requirements, but might be having a little trouble with an LF match. The HF detail problem is probably now close to being corrected. It wasn't trivial to fix, or come close to fixing. I am about ready to privately demo this correction. The bass is closer to similar than before, but I am not quite sure. The FA encoding process produces LF distortion, and we definitely do not want to reproduce any kind of distortion, including LF. This LF distortion in FA recordings makes judging the comparisons a bit challenging. Clarifying the 'lower HF' problem. Think of it like this: casually, the highs are mildly too strong, but the subtle details in instruments (trumpet kinds of sound) seem to be attenuated. When decoding, the sound of a trumpet should have more modulation in the lower-middle HF, but less of the 'itchy' kind of HF, kind of less of the edgy fuzz. The decoded result (in the experimental version) now shows that any of the higher registers should be waveform-wise are totally synchronized with the lower registers. FA tends to 'spread' the HF time-wise, causing it to be less correlated sounding, FA splatters the HF range of the trumpet sound across some number of low-milliseconds, and doesn't really sound like a trumpet at all. It is good to get the feedback, both the initial private comments and the public (embarrassing) results like this version. I thought that the decoder was in better shape when doing the release, but the myriad of types of subjective commentary, and the lack of almost objective comments has made this a challenge. The most frustrating issue is that I hear 'well' perhaps 20% of the time now, and never know when my hearing is accurate. Also, I had earlier rejected some misguided, perhaps well-intentioned objective comments, making the assumption that the comments were not well intended. The tools available are poor at measuring this challenge -- the reason why it has taken so long. The comments, still coming forward, help to re-baseline the results. To some degree, we all know the goals. I am pretty sure that the missing HF modulation problem will be resolved, and I hope that the bass will be correct. My single FA/never-FA recording example doesn't have enough variation to be an adequate example for comparisons. As mentioned above, the missing HF modulation problem is probably solved, or close to being solved. I am not sure about the LF tonality matter yet. John

Already, just got some feedback that the apparent freq response is biased towards thelower highs. This is very useful feedback, and was one of my original worries for this release. Earlier feedback on 700D, the previous release attempt, kept private, was that the decoding was a bit dead, and I might have been a bit too aggressive to increase the HF dynamics to be a bit too strong. I am going to be studying the matter before the 'final' demo try around Saturday. I will still be working on the best solution, which will not be to change the base HF freq response, but to change the HF dynamics. Just verified the base freq response for the highs, and if the nominal value would be '0dB' at 3kHz, then the following response would be 4.5kHz -0.3dB, 6kHz 0dB, 9kHz, +0.3dB, 12kHz +0.3dB, 15kHz - -0.3dB, 18kHz, -0.3dB. The entire range between 100Hz to 3kHz is 'flat' within +-0.05dB and below 100Hz has a rising response by a max of 1dB at 20Hz. The HF boost shouldn't seem to be prominent, but instead should appear to have a slightest HF edginess. It is impossible to fully describe the HF dynamics, but it is very possible that the necessary rising characteristics were implemented as too strong. Imagine the nightmare when working on this project with profoundly varying hearing with a small decline over time!!! I wish that I could easily judge the error before doing the release, but thank goodness that the very early reviewer/friend was able to point me in the right direction because it might have had worse problems. The correction will be well considered. I am still open to more criticisms. John

Good news... The demo is ready, and will be moving forward on other parts of the project after a few days of review... This demo release is probably IT, other than some edge-tweaks. These results are many light-years beyond any previous version, reminding misleading 'high quality' of the highly tweaked and totally insufficient original demos from approx 5yrs ago. These new decoding results provide very profound noise reduction and also less 'generation loss' as caused by the IP protecting FA encoding. Location of DEMO version '800D': https://www.dropbox.com/scl/fo/43tpy3s9jlucmd7ef3vx7/AFhd2nmZK13lnyzNUbdlIsc?rlkey=k3jwe3gbaxi3yetfus5umq14q&st=u51nbgjs&dl=0 (The normal demos are in the demos subdir) Previously, other parts of this project had necessarily been deferred because the decoding results were either of inadequate quality as a regression or insufficient improvement. Recently, with the original discriminator, the results were, more often than not, worse than the original FA version. (Expectation bias did distort my opinions -- too wrapped up in this stuff.) Thank you very much to the kind and persistent reviewer on AS whom has actively offered some participation, generally helpful & supportive along with his catching an unpleasant quality problem at the last minute on '700D' (literally not much longer than 12Hrs ago.) A few worrisome issues: Highs might be slightly too strong, sibilance might sometimes be worse than expected, once in a long, long time: a bit of fuzz on 'edges', once in a long time: weak birdie. Pleasant, not so worrisome results: Fantastic, fully functioning, hiss reduction on older recordings. very nice, improved stereo image. great definition. There will be a few noticeable quality improvements coming soon, but the decoder is on the tightrope edge of being truly 'correct'. In a few days, hopefully the project will be moving forward. For the next few days, gathering comments/feedback is still important, so will be tied up in the associated cleanup, perhaps even removing old/dead lines of source code. I am literally afraid of Windows and the troubles that it can cause for developers, being the reason no work has been done on the port yet. With the decoder on solid ground, I'll be able to focus on the myriad of potentially painful troubles in dealing with Windows. These new results were possible ONLY because of the new discriminator structure, that changes the discriminators into a 'for purpose' subroutine. The new discriminator is an order of magnitude more powerful than the old type. The design is capable of more accuracy, reaching the ideal as a 'null' instead of only a weak attempt of approaching the ideal. The design isn't as 'simple' as the previous, but works much better. Importantly, this is the very first, complete version with the new discriminator, and there might still be some helpful tweaks (e.g. multiplying parameters by 0.9998 or some such) for the 'release'. *I HELD ONTO THE OLD DISCRIMINATOR DESIGN BECAUSE OF FEAR, but ended up wasting a lot of time.* Further comments by me aren't really important right now, other than a suggestion for interested parties to listen. Some prefer to use the A/B/A/B snippet comparisons, others prefer to listen to the FLAC snippets. Unfortunately, the entire tracks cannot be publically demonstrated, but most examples do provide a reasonable comparison. SOME examples are too short for a given recording, perhaps a dozen or so of comparisons SHOW NOTHING. The A/B/A/B snippets are in the fscmp12, fscmp20 and fscmp35 sub directories, where the time-lengths are encoded into the directory name. Each file consists of the two versions of each track, in an A/B/A/B sequence. The first in the sequence is the 'decoded' (A), then the 'raw' (B). The 'raw' is preceded by the lower freq tone, and the 'decoded' version is preceded by the higher freq tone. The '95 demos' reside in the 'demos' subdirectory below the link above. SUBDIRECTORIES flacsnippets: Snippets of each track. Nothing else. fscmp12: Sequence of 12 seconds each of 'decoded/raw/decoded/raw' fscmp20: Sequence of 20 seconds each of 'decoded/raw/decoded/raw' fscmp35: Sequence of 35 seconds each of 'decoded/raw/decoded/raw' More info will be coming soon. Most importantly, the decoder really works and shows actual descrambler activity. John

Status: Probably technically 'perfect' coming in a few days or so. The reviewers don't have it yet, and will, of course, make the demos available ASAP. Lots of little corrections, including a descrambler about 10X more powerful. The source code change was generally small, just a better understanding of what is needed. It takes a HUGE amount of intuition to understand something as complex as the decoder. On the discriminators, I tried every plausible combination on the improved design -- the correct & best result appears to be the first version that I tried. The general design seems now to be easily defendable, of course, the source code quality is cr*p. My only *minor* concern is that the 'birdies' aren't fully under control. There is still room for 'tweaking', especially since the infrastructure is solid, and there are likely to be minor problems (like the occasional 'birdie') Nothing will be publically available until Wednesday at the earliest. If there are ANY delays, there will be a short list of demos instead, but at this point, the sound is so very good and I don't expect any delays. Examples: ABBA Gold (More Gold) essentially perfect, with some reveals because of the detail. Karen Carpenter sounds the best that I have heard since the 1970s. I am not a classical expert, but I can clearly image the location of the instruments. I know that I have been 'crying wolf' for years. This time, the 'wolf' has really gotten me. I wish, I wish, I wish that the demos were ready now!!! Sincerely and relieved... John

Status: minor update... The descrambler portion of the decoder is still being worked on. There has been LOTS of learning, and very importantly LOTS of ferreting-out various configuration parameters. Just now, for example, another minor error in one of the parameters has needed to be corrected... (a change in an obscure parameter from sqrt(2) to the arcane 2.213, rather major difference) The resulting offset frequency for the sidebands (used for demodulation at each step) was erroneous, and reduced the expansion and degarbling performance of the descrambler. The errors weren't so bad that the descrambler was left to be totally dysfunctional, but instead didn't function as well as it should. So far, there are still very few non-standard values for settings (e.g. gains are mostly like +-3dB, etc.) The center frequencies for each descrambler step (each discriminator) are still the same and still appear to be correct. Being able to avoid many 'tweaked' parameter values does bode well for correctness. When I am in contact with each of the reviewers/contributors, I tell them that I will try harder to avoid making questionable demo releases. When there is a release of a full set of demos, it will be fully functional per my perception -- and public claims about a release will be made only after at least one reviewer telling me that the result is 'okay'. Once the performance is close to ready, there will be a short list of examples made available. I am intending that a 'short list' of demos will be made perhaps 1wk before a full set of demos. That 'short list' might not be 'full quality' yet, but I still want to include interested parties in the process. My previous misguided time estimates are regrettable, and I do continue to apologize. The descrambler is more complex than what I could even think to imagine, but that is not an excuse for the errors in my time estimates. The descrambler *IS* functioning, and the decoder *IS* working, but not to the level of my expectations. Depending on recording, the signal clarification is between barely noticeable all the way up to significant quality improvement. I am hearing improvements, but not profound enough. Obviously, I am still not happy with the results, and I am resisting the temptation to 'throw it at the wall and see if it sticks'. When I am VERY sure that most people will hear at least some improvement with no significant quality regressions, then the project will be ready to demo. My own standards have increased, resulting from a learning process and better understanding what audiophiles *generally* expect. I still have the attitude that I AM STILL LEARNING, and much more open to 'input' than earlier in the project. So, bottom line -- significant progress is being made, some health issues have required more careful/deliberate processes, and also won't do a release beyond 'short list of demos' until my new, more exacting standards are met. I'll do an update in a few weeks again, hopefully with at least a 'short list' of demos. John

Status: There is a version, V21EV757, that could be demoed right now, but deferring due to caution. It has one flaw that I can detect, the highs under certain circumstances are attenuated just a little more than they should be (IMO.) There is a very narrow balance between a stealthy kind of gating, esp at the HF vs. the opposite. It is possible that '757' is correct, but I want to double-check, and correct if needed. The new descrambler is able to do a lot more processing, including viciously overshooting the goal. Overshooting the processing goal is usually far worse than leaving the FA (raw) recording alone. The sound is clean, tight and sometimes intense (when it should be intense.) The defects of the older versions should be long gone. It would be sad to introduce a new defect, and that is why the '757' demo and soon-to-follow decoder binary/source is being deferred for approx 1 wk. If, any time this upcoming week, that '757' is found to be 'IT', or the yet non-existent 'updated' version is correct, then the demo will be made available ASAP. Good news on the technological front: the new, 'active' discriminators/descrambler are a big success. Many flaws, including the processing being far too weak, are corrected. As mentioned above, none-of-us wants any new flaws, and that is why the demo is being deferred for up to a week. If there are requests for the '757' demos, I'll be happy to grab a few of the already complete demos and 'show them off'. Even ABBA Gold sounds good, including reproducing the more difficult tracks. 'Dreamworld' is still somewhat problematic, but without an intact 'pilot', Dreamworld will always be 'trouble'. (The pilot, sidebands from the original scrambling process, is easily destroyed by the wrong kind of <20Hz rolloff.) John

Thanks about the nice comments about getting over the illness. Doing very well, and been home for a few days. The previous descrambler was based on a simplified design, think of it as 'passive'. I had pushed the 'passive' concept so far as to the work being 'frustrated' instead of 'productive'. During the time-off, I had realized that pushing the old design was just a waste of time. The new design is more of an 'active' concept, which has more pitfalls, will do better than just approaching 'correctness', but has a risk that it can overshoot the goal. I apologize for all my 'optimistic' evaluations in the past, but admittedly there were some biases in my own reviews, where the very slight changes or improvements sounded more impactful to me than to other people. The previous 'improvements' ended up still leaving too many flaws in the sound. The frustrated 'improvements' mostly just 'wobbled' around the best possible result. It is very, very important to realize that the descrambler is intended to be just a descrambler stage of the decoder. It appears that tuning the descrambler to do a lot of 'expansion' is not a correct application of the design. The descrambler naturally does a small, natural amount of 'expansion', but the primary purpose is to clean up garble in complex mixes of sources (e.g., vocal chorus type mixes.) With near-'virgin' relatively un-accommodated hearing, the clarification is obvious, even though in some cases the sound might be a little less satisfying. The clarified vocal mixes seem to produce more easily resolved actual detail, but the 'descrambled' sound is less busy, being less 'busy', the sound just might be less satisfying. There is opportunity to further review this 'less satisfying' sound, and truly determine if it is correct, or perhaps there might be actual lost detail. I will try my best to avoid publically claiming near finality until the improvement is close to the best possible. So far, in the last few days, the more powerful design produces better results than any previous version has done. I WILL be mentioning and even demoing interim versions, but I'll label them as interim, and try to avoid major proclamations until the claims are substantiated. Also, I'll spend some time on creating a Windows version of the decoder. The Windows decoder has been of lower priority because the functioning has been a stronger focus. The functioning is much better, so I'll start making some progress, hopefully with results coming within a few weeks. (I have no independent Windows licenses, and the laptop gifted by my Sister might or might not function, therefore there might be a few more necessary steps than just installing the development environment.) NO MATTER WHAT, the current decoder is effective at 'denoising' many older FA recordings. Much of the time, the denoising is without obvious artifacts, and alone might justify using the decoder when the hiss is profound. Therefore, the Windows version might still be helpful from time to time, even if the descrambler isn't fully functioning. (Examples of recordings where improvement might be worthwhile include: Brubeck's Take 5, Getz's Girl from Ipanema, some of the early Carpenter's albums, etc. The Brubeck album has especially heinous amounts of hiss that appears to be almost entirely removed, without substantially removing detail (at all?).) With the new descrambler, and a little more deliberate work instead of being panicked on the old design. There should be more useful 'descrambler' based results coming before the end of this decade…😁 As mentioned above, the descrambler is 'descrambling' very effectively and isn't subtle. However, not sure if it is correct, yet. Thanks, and there is a new sense of hope. John

Was working on the release, got hospitalized for a frustrating illness. I won't be up to snuff for a week and then be another week of work on the decoder. Recent reviews showed that the amount of processing was still a little bit too mild. I won't be online much much for the next week. The voice recognition on the cell phone doesn't work as well as typing. The project will never stop until it's finished. But like recent history, it will take longer than any of us hope. Thanks for your continued interest!

Status: You probably realize that I am currently having real troubles finding 'correctness' to my satisfaction. Reaching the goal has been elusive because of the few never-FA A/B references available. The only reasonably good A/B is a Sheffield Labs recording, but even with a 'perfect' match, it isn't enough. My other 'never FA' recordings have been polluted by intrusive mastering. For a reference of the general character of the sound, some very high quality true master tapes have been used (never FA, direct recording.) Doing the comparison at this level, it is obvious to me that the earlier versions of the decoder were missing a 'certain something', which I have found a good approximation, maybe even a good emulation. * all the 'technical' aspects have the known errors zeroed out, so the 'EQ' or specific processing parameters are still not 100% correct (or might be correct?.) The sound should be pretty good!!! The decoder sounds almost impossibly different and probably better than previous versions. I'll be presenting the current best version branch that 'sounds good', and 'close to correct'. I wish I could do better right now, but with all the variables, I am going in circles. The technology is correct, but the final settings and minor add-ons has been frustrating. A public review of the current 'sounds good' version seems to be a 'good idea'. If all goes well, the first complete release with a Windows version will be based directly on the upcoming demo. First, I am limiting the first level reviewers to 2, maybe 3, not to exclude anyone. To avoid excluding anyone, the copies will be available publically and privately as usual. The private versions being more complete. I do need to limit the scale, because right now for personal reasons. I need to keep things simple. Known reviewers will be asked for their help until I can find a few still interested. Recently, I haven't been communicating very much (to anyone.) With the mods based on the sound of the master tapes, and an attempt to match all the pop/orchestral FA in my archives, the sound should be perceived as pretty good. If I have 'missed the boat' on the emulation of never-FA recordings, then there are a few more possibilities ready to try if the 'sounds good' version doesn't 'sound very good'. Right now, the results of decoding sound *really really nice*, but might still be a few days, probably upcoming Monday?, yet to make the final version. This time a demo release *WILL* be a few days, but won't be available until I have reviewed the demos myself with happy results, which does take a lot of time. (Most people only check a few of the recordings, but I tend to listen to approx 1/3 to 1/2.) The 'correct' version is still coming, once I can find good never-FA without ham-handed mastering. Also, the work might be easier after a several week slowdown. Sincerely, and fairly happy with the sound and perhaps reasonable approximation to 'correct' (or maybe not.) John

Status: Been getting questions about the lack of recent postings. The fewer postings do NOT mean that the project has slowed down, and this location on AS is still the primary location where status is mentioned. The reason for the slowdown is twofold, first being the decoder being 'complete', but still trying to get the perfect match to the few available never-FA/FA and also a proper guesstimate about correctness on other recordings. There is so little to talk about at this point, and demos would be almost a waste of time for feedback, because the results are so close to correct, a verbal description just might be confusing. From the techno babble standpoint, there are so many little mistakes that talking about each one (perhaps 3-4) would be meaningless to you, but also the fine details elsewhere might be 'too many words'. There might be two major 'techno babble' type corrections: the discriminators had an incorrect set of filters, and the Hilbert transforms in the antifog had a hideous error in the FIR coefficient calculations. Both are straight forward and don't really need many words to describe. Suffice to say, there is a profound subjective improvement with a correctly functioning antifog and more correct and linear discriminators. Both corrections together produce rather pleasing results. Alot more details are available upon request. There is a subjective weirdness that seems to be sonically transporting me back to the late 1970s… This effect hasn't manifested on earlier versions of the decoder, and certainly not on the FA recordings. The sound is NOT an ambience, but almost opposite, where there is a clean stability in the sound. I sure hope that this effect shows a better 'correctness' rather than a 'sound-defect'. I like the sound, but if it is found to be *incorrect*, then the subjective effect will not be missed. (The differences in the descrambler/discriminator are small, and easily separated out -- there is NO NEW processing, maybe just a better implementation.) I hope that the decoder is coming soon. There will be no known bugs, or at least, only very minor bugs in the next version. It is taking time and is very laborious to listen to the same songs OVER AND OVER again, but it is necessary for the most completely possible Q/C. Demos are always available, either from my library or selected from your own. Demos will be publically available, there WILL be demos, most likely soon (hopefully days, not weeks.) It is getting closer and closer. John

Status: At this 'very close to release' timeframe, there have been some frustrating stumbling blocks, one of which has been resolved in the last day. A lot of these stumbling blocks are small, relative to the complexity of the algorithms, but still cause critical problems with the sound quality. One bit of trouble has been a 'rough' sound on certain mixed vocals, e.g., approx 80 seconds in to Steely Dan Aja (S/T recording.) To me, the rough sound has been unacceptable, and I have promised to myself and suggested strongly to others that there would be NO known bugs in the next version. This means that the release needed to be delayed. A side benefit of THIS correction is that some peripherally associated quality improvements have been added. About 'the fix': I have this mental lock-in of 'the rules', where 'the rules' associated with the design must be followed, otherwise the exception to 'the rules' need special consideration OR 'the rules' must change. In this case, one of 'the rules' needed to change. Forever and ever, I locked in the Q values for the EQs to be 0.0 (meaning first order), 1/sqrt(3) (degenerate Bessel), 1/sqrt(2) (degenerate Butterworth), 1/sqrt(sqrt(2)) (degenerate Chebyshev), along with sqrt(2), sqrt(sqrt(2)), and in the discriminator, a Q value that is exactly or approximates '20.0'. This list of Q values has been very useful in maintaining sanity in the design, while likely matching the original design values. One of these Q values was wrong in most of the critical situations: Q=0.577 (1/sqrt(3)) degenerate Bessel appears to have often been incorrect. The correct value in those critical cases appears to be Q=0.50. Why is this little detail being mentioned? Because in lieu of doing a reckless, sloppy demo, I am searching out every bit of distortion in the sound. Unfortunately, objective measurements are difficult to interpret, so are dependent on my questionable subjective reviews. I DO have troubles with freq response balance, but am also taking great care using multiple listening transducers to avoid all biases that can be avoided. Finding the source of the 'roughness' in several vocals has taken two weeks or a few months, depending on how you look at the definition of the problem statement. These last bugs have been tedious to find & correct, sometimes causing valid complaints from those who 'just want the decoder to work'. The 'decoder' is not a 'remastering hack' piece of software, even though it might have seemed like it was at first. As mentioned before, I really thought that this project would be much easier than it really was. Without using a lot more 'useless words', PLEASE just accept the idea that IF there are obvious bugs left in the decoder, then all of this time that I spent (and others had helped) would be wasted. PLEASE be patient, and sometimes my over-enthusiasm is the only thing keeping this project going. My over-focus and over-enthusiasm has had other negative effects on the project, but it wouldn't exist without almost impossible patience and persistence. It would be a terrible waste to quit now, when the only bugs are almost provably like the 'Bessel' vs 0.50 Q value bug talked about at the first of this message. Even more sad would be to represent an inferior result as being the same as the correct ultimate goal. It is important to try HARD to keep this from happening. *Personal demos are available (just tell me in PM or email) at any time*, but remember there are still bugs, hopefully now gone? Maybe this IS the last bug? Probably not, but just might be the last 'significant' bug? The search is continuing for at least the next day or so. The specific detail about the correction is NOT significant, but the fact that any time there is a 'bug' in the audio, the matter IS being resolved. If/when the release (demo first) is REALLY produced, it will happen after practically all bugs are addressed. On the other hand, *SOON* THERE WILL BE A FEW SHORT DEMOS, perhaps today/tomorrow or so. The only reason why the short demo hadn't already been made today or yesterday IS exactly this vocal roughness issue, and a previously recognized critical complaint that the processing is too mild. (I have already staged the 'short demos' twice, but not 100% happy yet.) So very sincerely, John

Status: Project still moving forward, stuck in the 'Russian Doll' syndrome, or perhaps layers of onion... (The length of this message is regrettable, but without reading this whole thing, the decoder, as a *decoder* is probably better than might have ever been expected after knowing the challenges.) --- The decoding process is likely more complex than the encoding side of things. Some aspects of precise 'decoding' appear to be technically impossible. Because of having some amount of integrity, it is starting to make me 'shudder' describing the project as a 'decoder', because decoding is nearly impossible. However, the 'decoder' is probably coming as close to complete as possible. As a matter of 'shorthand', I'll still call the project a 'decoder', but admit to the limitations. The most recent layer of the 'onion' is not a matter of being structurally correct, but instead the specific parameters needed, and how wide the discriminator detection curves must be linearized, etc. Some of this is quite esoteric, but the project coming along nicely. In the project, most 'parameters' are ending up being values like sqrt(2), all kinds of precise conversions and happily, most 'fudge factors' are numbers like '2.00000'. There are still some compatibility tweaks, but very recently the '0.9999' and '0.9998' type magic numbers have gone away because the rest of the system is becoming very precise and produces repeatable results. There are still very few uses of 'strange' numbers. Several years ago, then several months ago, and even several weeks ago, it seemed, to me, that the 'final version' was very close to being ready. Unfortunately, my expectation and experimental biases encouraged excess enthusiasm. These biases were also amplified by very unexpected variability of my hearing perception and my naive expectations about headphone performance. Nowadays, most of these problems are less important, because the parts that are dependent on my hearing frequency response balances, etc are now diminished. It is still futile to objectively measure many of the behaviors of the decoder, but recently the A/B choices have been more obvious as the processing becomes more aggressive. (Since the decoder is now approaching a practical goal, any settings that had been 'too conservative' are now configured more realistically.) Some still-necessary subjective comparisons are still troublesome where the source of an error might be 'hidden', I have wasted a lot of time trying to fix bugs that actually reside somewhere else. Most comparisons are searching for overt elements of distortion, and NEVER just about 'sounding good'. After some discussions with one of the reviewers, to avoid disappointment, eventually it might be helpful to add a limited', but very high quality automatic-remastering'l. 'Remastering' is best not attempted until the recording is 'fully decoded'. I have some examples of totally messy commercial remastering examples based on the FA distribution version and NOT on the pre-FA original. It seems like it would be best to avoid creating worse messes than many consumer recordings already represent. The decoder must be reasonably fully functional before a simple remastering phase is implemented -- and I already have some good ideas about a simplified remastering phase that should work nicely. (The 'add-on' remastering is like creating a simpler & higher quality FA again, but without the worst problems and without adding the hiss and grain back in. -- a less aggressive FA with the good parts only.) Most importantly for now, the goal of the project is still focused on 'recovery', not 'mastering for subjective experience'. In lieu of a 'final version' right away, in the next week or before, there'll be some demos that show 'improvements'. As a matter of integrity, the demos will not be 'cherry picked', but will accurately and obviously demonstrate the effects of 'decoding'. Along with the 'subjective' examples, also plan to produce some examples showing the effects of the 'pilot' handling. The 'without Pilot' will simply result from a soft LF rolloff below approximately 15Hz. Depending on recording, the loss of the pilot portion of a recording can be noticeable and produce a loss of quality. Without the 'pilot' left in the recordings, the decoder project would likely have only been a curiosity or failure. The 'pilot' is a key portion of the decoding process. Earlier versions of the decoder showed over-enhancement because of vain attempts of 'decoding without the pilot'. Grossly increasing the dynamics without the 'pilot' hint mostly causes a loss of quality. Sadly, vinyl sourced and mp3 type recordings don't have much (any) useful 'pilot' information left over. Recordings made with perhaps a single generation of tape or no generations between the FA creation and making the distribution itself appear to be decodable. For example, one of my major nemesis 'ABBA Dreamworld' appears to be a diminished quality tape copy of the FA signal, therefore some dynamics are lost, and also creates some distortion in the sound. (Some loss of the 'pilot' signal strength doesn't appear to be fatal, the 'pilot' seems to have some characteristics of being 'FM'. For good quality, any <20Hz rolloff must be avoided. The pilot appears to be intact on most digital distributions like CDs and other similar distributions that haven't been lossy-compressed. Previously, I have suggested that these messages won't be as frequent as before, and that will continue to be true. There is little risk of the project stopping, so just because there hasn't been a message for a week or so doesn't mean that the project is stopped. The content of a lot of potential messages would be overly esoteric and require too much context. Even the mention of the 'pilot' is bordering on 'TMI', but the 'pilot' does create some obvious changes in 'decoding' results. My time estimates are almost always inaccurate, but the 'pilot' example is expected in a few days. I am hoping that a few subjective examples might also be demoed. Very successful improvements could be demoed *right now*, but it is best to demo something that is *accurate* and not just *for effect*. It seems like true accuracy is still indeterminant, even though there are few serious regressions (if any.) Nowadays, my worry is more about the decoded results only seeming to be 'clean', instead of 'surprisingly good'. The 'surprisingly good' subgoal might require a 'remastering phase' that brightens the recording intelligently, without excessive FA damage. This 'worry' is still being reviewed. (still peeling the onion, but close to the core.) John

Status: Project still progressing. I keep thinking that the decoder is 'finished', yet find ways to improve it. All focus for the last couple of weeks has been on the descrambler, which has now become a rather powerful expander (probably also compressor). THe current configuration with the decoder provides too much expansion but happily without artifacts. The goal over the next few days -- find the correct setting, which is precisely two knobs (settings). One knob generally focuses on the 'getting louder' and the other is 'pushing down the volume', both are instantaneous. There is absolutely no surging, and given reasonable settings, no distortion that I can hear. When working in conjunction with the rest of the decoder, it generally works very well -- super well, without audible distortion that I can hear. It seems to adapt well to most consumer recordings. The current failing is that the setting for the correct amount of expansion (descrambler based decoding) isn't easy to determine. The correct setting just might be 'neutral' or might be an arbitrary number. The current setting is too strong, but without an obvious 'tell'. In recent days, the descrambler is no longer a special purpose concept, but instead can be adapted to be the 'impossible expander'. I am hoping to produce some plausibly correct demos soon. In the last few hours, a run of ABBA Gold produced results with essentially no distortion, but with dynamics that make the recordings almost unlistenable. This excessive expansion is NOT the HF 'ear burning' kind of result, but is simply 'normal' results with too much subjectively flat expansion. After the failed attempts at a release, it seemed like a more repeatible and more stable result would be achieved by processing that could be precisely controlled. The new descrambler,, totally based on the previous, but with more controlled design, will give a much better chance of a release that will succeed. I have decided NOT to distribute the snippets for the recently create ABBA Gold results even though the results do reside in the private Dropbox area for now. Distributing the snippets would be confusing because they are known to be faulty, mostly just of deeply technical interest. I doubt that anyone will ask, but if asked, I will happily make the snippet version of the incorrectly strong dynamics version of ABBA Gold available. Frankly, it isn't worth it to anyone except those interested in the *descrambler* instead of the *decoder*. Mostly, suffice to say, progress is still being made. Trying desperately to avoid any more failed releases. After the decoder is truly released, anyone wanting to make a 'super expander' might want to contact me for source code. At that point, I'll have time to split the source to be a reasonably separate C++ class. John

Status: I apologize for the ongoing delay. But, there is really good news... The pilot processing is working amazingly well, and there is very noticeable improvement in distortion on difficult material (e.g. Bridge over Troubled Waters.) I am doing massive A/B comparisons relative to earlier versions, relative to the FA originals, and importantly relative to the few never-FA recordings that I have. Unfortunately, some of the never-FA recordings had been manually mastered, some even had reverb added -- yikes. The most frustrating are some selections of vinyl-rips that come from damaged vinyl and not-so-good phono preamps/cartridges. Most of the never-FA require repeated tests and use of interpretive best-judgement. In ALL cases, the typical failings of the decoder are being scrutinized, and when possible, eliminated. There have been bugfixes that are very important for accuracy, including correcting some 'ghosts' that I previously had been chasing and making choices based on those ghosts. I believe that all have been remedied with ongoing day-to-day skeptical reviews (by me.) I have every motivation to remove any bugs, whether or not I like the sound. * The smallest selection of reviewers will soon be presented with demos. THESE demos will also be made public in truncated form. However, something is coming soon, almost happened today, almost happened several days ago, almost happened a few weeks ago... Each of these delays has intended to AVOID WASTING PEOPLES IMPORTANT TIME. The important correction/improvement in descrambler pilot handling has been super-critical for the best results. The pilot has now become very powerful rather than just an adjunct. In fact, the pilot is so powerful that careful A/B is needed to avoid 'enhancement' or too much improvement. Pilot enabled 'enhancement' is sometimes difficult to detect, but comes the close to best 'transparent enhancement' that I have ever heard (including fancy FFT dynamics processing schemes.) The infinitely fast attack/release along with the hints coming from the 'pilot' produce 'interesting' results. Frustratingly, for THIS project, use of 'pilot' for enhancement is not allowed. The only valid use for the pilot is for 'correction'. Previously, I have claimed 'just a few days'. Every time I start working on final review, I find more and more ways to improve the decoder. Therefore, public results might be tomorrow or might be a week from now. Quality is more important than delivery time, no matter how much the delays might seem to be unacceptable right now. The goal is 'practically perfect' and not just 'sounds good'. This goal is almost impossible to achieve, but will be achieved SOONER THAN LATER. I have my own limitations that make 'sooner than later' very important to produce results in a few months. (Nothing dire, just personal matters.) John

status: Making real progress. The nuggets of feedback have helped to recognize that some bugs that crept in. The decoder has been *working*, but with a qualification that the processing was generally correct, but the scale factors drifted. These errors almost totaly broke the discriminators. Breaking the discriminators is about the same as replacing a 100HP engine with a 1HP motorized bicycle engine. The processing worked, but was 'under powered'. The good news is that a cause of 'limited improvement' has now been found. There was continued feedback that suggested that 'more was expected', and finally 'more improvement is coming'. If all works well, I am able to work today and tomorrow on the decoder, then an upload tomorrow night might be possible. Since this improvement is *profound*, there might be more secondary issues that need resolution. The associated drift corrections might cause more delay, but I will really try to make something good available as soon as possible. John * By far, most of the recent bugfixes are still 100% applicable to the discriminator 'oops'. Therefore, set-up of the driving functions for the discriminators,etc. are still functioning, and the corrected discriminators appear to be functioning surprisingly well.