Jump to content

John Dyson

  • Content Count

  • Joined

  • Last visited

1 Follower

About John Dyson

  • Rank
    Audio DSP/SW developer, sometimes listener

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. Richard didn't like the LF frequency response (actually, I did the rolloff on purpose, still have an --hpf switch -- too much LF causes troubles.) Anyway, here is the new comparison between DolbyA and DHNRDS DA!!! :-).
  2. Good idea, something could be worked out for that... I gotta get my a** in gear to add in the feralA section soon (days, not weeks.) I think that there is a good set of standard formulas that haven't let me down. They even work for the odd recordings like 'Linda Ronstadt', where the feralA has some strange behaviors. Got something that I am actually very proud of. I just did a test decode of one of the master tapes in my library. Did a spectrum display of the snippet... Take a look at the difference between DolbyA and DHNRDS!!! (Yea, I know that the lows around 20Hz are different, but otherwise.) The audible difference on the snippet is qualitative and shows some improvement (I am so frustrated, but I'd lose all access to anything if I shared anything like that!!! I really want to brag... But take a look at the match of the spectrum. Sure, that isn't everything, and almost isn't anything -- but it does show that the general freq resp balance could be close.) Figure that the DHNRDS DA architecture is TOTALLY, ABSOLUTELY conceptually different than the DolbyA HW. It is better than I could wish for.
  3. Hmmm... Yea -- mega CPU usage for sure. I never said that the DHNRDS didn't suck up every last bit of my 4 core Haswell :-). One criterion was that at the highest quality level (or nearly the highest quality level), that the program run in real-time on my 4 core Haswell.... Today, the way that it is written, and the available modes, it can run between 3X faster than realtime to 2X slower than realtime. Of course, the 2X slower is an experimental mode to find asymtotically the highest quality that can be achived using the current techniques. The slowest practical mode is about 1.2X slower than realtime, but most of the time I play and run tests at realtime while listening. (So I decode while listening, just like it was a compressor or expander that can run in realtime.) It is very easy to make the decoder run perhaps 4X faster than realtime, but at that quality level it would be roughly equivalent to a true DolbyA (probably squeeze in a few quality improving features at that speed.) Providing a service like that, if it ever became popular (unlikely), it would be good to modify the program to run in the cloud :-). John
  4. I really wish that I could play a master tape to demo what actual, real material sounds like. In fact, I have some of the groups in the 2nd set of tidal examples. Without DolbyA decoding, of course, cannot play the tapes without getting sound even worse than feralA. I have direct A/B capabilitites of DHNRDS vs DolbyA, and a normal DolbyA master tape can lose serious detail (percussion/transients like normal, actual instruments.) When there is no serious detail to be lost, then they both sound practically identical (really.) One, easy to hear example is that a cymbal crash with background high-hat and bass guitar, on the DHNRDS the background high hat and bass guitar stay the same, correct level. However, with the DolbyA, it is almost like dynamic range compression, where the bakground high hat and guitar levels are instantaneously dropped, then increase back to the correct level. Some day, the labels will actually care about their quality, instead of just talk about it. John
  5. Thanks for your feedback -- I like doing nice things for friend/correspondents and also for the music. Trying to decode for the general public, that is 'professional' feralA decoding is much more difficult, than just making it sound good. I have been trying to keep my standards high, but there is still too much 'experimentation' and depending on hearing audible defects. I am very, very close to understanding the filter set needed for feralA decoding. The DolbyA decoding done by the DA decoder is probably more perfect than a true DolbyA, but the feralA decoding has a couple of complications. One is that the person who decodes must be able to guess at the filter configuration, and then also know the sound of decoding defects. If I was decoding just for me, it would be easy -- sounding good is actually pretty easy. The goal is to 'sound good for everyone', and that is very, very hard. It is enjoyable to make people happy, and hopefully increase the interest in the older music. The feralA decoder should make it much more practical for the advanced audiophile to correct the recordings for themselves. While doing these feral decodes, I am learning what is needed for creating the feralA filters. The feralA decoder configuration will NOT be commercial, and will be a useful subset of the professional DolbyA DHNRDS decoder. Still have to work out the logistical details. Anytime there is interesting music, music that you love (anyone reading this), and you think that it is feralA -- just let me know, and maybe I can do something with it for you. (That is do a decode for you.) Later on, the decoder will be more practical to use, and less of a science project to use it. At that time, then the audiophile can give it a try themselves. I'll probably be around for a long time to 'give a hand' if needed though. John
  6. Take a look also at the Rumors8 versioin. It isn't 88.2/24 yet, but might even be better (tradeoffs between each version -- tryign to take the best of the early and the best of the almost-CD sound.) I'll send you the information privately. If you really like it, I'll make the 88.2/24 available. About the signal levels -- I kept the same levels as on the original recording. That is, I didn't normalize each song. I have SOMETIMES found that if each song is normalized, then the perceptual levels don't match (even though the technical levels are all at -1.0dB peak, the perception is that the levels vary too much.) So, to be conservative, I just kept the natural level that came out of the decoder. That means that some songs might even be -10dB peak or lower. If I really focused, I could probably do better to match the perceptual levels than just usnig whatever level they come out after decoding. John
  7. Wow -- that little tidbit about Carly Simon's recording desires helps me to understand why some of her recordings are so difficult to make them sound really good. There is an edginess in her vocals, some kind of manipulation around 7-9.5kHz or so -- all I know is something is wrong, and cannot put my finger on the cause of damage. A spectogram shows that something was done oddly in the 7-9kHz range. Her vocals can be cleaned up to some extent, but that 'distortion' takes away some quality. I understand that she might have wanted her vocals brought out more strongly -- but please NOT LIKE HOW THEY DID. I did find that cancelling the 8k through 9kHz band in the the 'stereo difference' channel (the M of M+S) helps noticeably. John
  8. Hey everyone -- got some good news!!! Bottom line, I am very very close to the FeralA decoder design... Previously (until this Rumors fiasco, I have to realize that community efforts stress me out too much), I had a rather contorted, but well thought-out set of filters that worked better than anything previous. When having troubles getting the exact sound that I wanted (it might still need a minor pre-decoding tweak), I tried a different, simpler set of filters. Voila -- it looks like the shape is correct, just maybe the values might still be a bit on the 'hot' side. I tried the new, simplified filter concept (today) on Carpenters, Brasil'66, Herb Alpert & Tijuana Brass, ABBA, etc.... They all work better than ever!!! The beauty of the new filter concept is that there are fewer modes, and mostly just a set of three parameters (I hope.) Parameter 1: the MF compensation Parameter 2: the lower HF compensation (3k through 6k) or (4.5k through 6k) Parameter 3: the upper HF compensation (9k through 12k, maybe 15k) Each of the parameters is one number... Parameter 1 is a 'mode', of a set of 6 possible modes, most likely only 2 are needed to try. Parameter 2: -1, -1.5, -2, -3 dB Parameter 3: -1, -1.5, -2, -3 dB There is a flag for Parameter 2 that specifies whether it is 3k -> 6k or 4.5k -> 6k Parameter 3 will almost always be 9k -> 12k, but 15k might be useful. The MF compensation is non-trivial, but it appears that there are few choices. One final mode would be a specific setting for each of 500Hz, 750Hz, 1kHz, 2250Hz, 2500Hz, 3kHz #1, 3kHz #2, 2750Hz, 4.25kHz, 4.5kHz, 5.75kHz, 6kHz, 8.75kHz, 9kHz, 11.75kHz, 12kHz. Some are Bass shelves, some are Treble, but they are always one or the other!!! Normally, the individual settings are not needed, and with the current propsal above, I would never need to use the individual filters as I noted above!!! ------------------------------------- Bottom line, ignoring the technical details, I can simplify the settings to the point where decoding will only need three command arguments for filters. There are a few other modes also -- but it now seems practical after being given a few normal examples!!! Before this, the filters were too complex!!! John
  9. Here are some good results from the album 'Fleetwood Mac', using the same formula as 'Rumors'. It appears to need approx the same formula!!! This are only snippets for now. John 03 - Blue Letter-snippet-dhnrds.mp3 04 - Rhiannon-snippet-dhnrds.mp3 08 - Landslide-snippet-dhnrds.mp3 10 - Sugar Daddy-snippet-dhnrds.mp3
  10. I use the term remastering as a 'choice by taste'. I try to avoid 'choices by taste' and reserve my own changes for technical corrections. Refer to my previous post about the highs not being supported by the midrange, therefore producing the previous weird sound. This manifestation doesn't happen 'after' decoding so much -- the expander gets confused by the missing match between highs and midrange. All of the recent changes have been to better match the highs and midrange. (The weird sound is very noticeable -- I was looking in the wrong place for the problem.) The technical corrections have a knifedge before decoding -- the problems are easy to hear IF LISTENING. I wasn't listening for that specific problem, but instead went down the rabbit hole of HF EQ before decoding. The problem wasn't really the HF EQ, even though that seemed like the problem (even to me.) It was a gouged out upper midrange!!! If the filters were specified, I wouldn't even have to make the filter choices at all... * On severe input errors like what I previoius did, A true DolbyA would go nuts and create an edgy, almost noisy beat between the channels -- seems like a kind of harshness or distortion. The DHNRDS says: I won't distort, but instead, I am going to give the listener an acid trip with weird sound instead :-). I should probably create an option to disable all of the anti-distortion mechanisms in the DHNRDS, but that would require literally 20-50 conditional 'IF' statements to nuke the avoidance methods. The worst of the problem was the wrong frequency choices for the upper midrange EQ correction. It is a well defined set of filters (I explained it elsewhere), and I simply made the wrong choice -- therefore causing a rabbit hole. I popped out of the rabbit hole and fixed the correct problem. (There are discrete frequencies for the filters -- must have been some kind of internal standardization for feral EQ.) John
  11. The last two most recent links (Gold dust woman and Never Going Back) are still valid, but have new files backing them. They are MUCH better now, the midrange now supports the highs correctly. These are full quality 88.2/24 bits and don't sound weird. Give these a try, I'll ahve to disappear them by the EOD today though. I added a third one... The PM group has all of them. (Like usual, I am embarassed by the previous ones -- after a friend shared snippets from the LP, I paniced. The LP had missing highs, so I had to rebalance to the point where the highs were just barely supported by the midrange -- it is a technical choice, not by taste. Such a correction MUST be done before decoding, or it won't really help. I could turn the highs down by perhaps another 0.25 to 0.5dB before decoding, but the numbers I used are the typical rounded values like 0.75, 1.0, 1.5, 3.0 dB steps. They seemed to have used such values for the EQ and is the reason why I have settled on it. The next step would be+- 0.75dB BEFORE decoding, which would be very severe, it makes HUGE changes in the sound -- sounds worse than 3dB error after decoding!!!) https://www.dropbox.com/s/0n0zi6iwp8d5omo/Fleetwood Mac - Rumours - 03. Never Going Back Again.flac?dl=0 https://www.dropbox.com/s/t903c51qb781htg/Fleetwood Mac - Rumours - 11. Gold Dust Woman.flac?dl=0 https://www.dropbox.com/s/mzxp393wpedv4cf/Fleetwood Mac - Rumours - 08. You Make Loving Fun.flac?dl=0 John
  12. Here is an example of a completed song... It might need one more iteration, but this is damned close to correct. It cannot get much better (in fact, a DolbyA HW cannot do this good.) This has detail that has probably not been exposed since the recording. YOU WILL HEAR RECORDING DEFECTS THAT HAVEN"T BEEN HEARD -- EVER!!! If anything, I might need ot do a slight cut of -1.0 dB at 14.75kHz and -1.0dB at 15kHz with Q=0.50, which will soften the highs a little. Earlier decode attempts had those pre-decoding EQ, but that EQ is needed before decoding, another Hr away!!! (The defect is that there is an 'edge' in the vocals -- and beleive it or not, a cut at 15kHz makes a HUGE difference, even in the vocal range -- it is all related to the behavior of DolbyA decoding.) EDIT: I decided to do ONE MORE attempt tonight with the 15kHz fix, it makes a REALLY GOOD improvement. https://www.dropbox.com/s/t903c51qb781htg/Fleetwood Mac - Rumours - 11. Gold Dust Woman.flac?dl=0 https://www.dropbox.com/s/0n0zi6iwp8d5omo/Fleetwood Mac - Rumours - 03. Never Going Back Again.flac?dl=0
  13. Enjoy what you have!!! Do give the new attempt a listen -- it is most likely better (but no guarantee!!!) The big difference is that I am not changing things dynamically, and being careful to FOLLOW RULES, and FIX BUGS, not do artistic modifications Sometimes material just sounds strange, and that needs 'fixing bugs', not doing complex tone control. A good example is stereo image. Believe it or not, the frequency response balance can change the stereo image. Sometimes things can be strange, yet fixed with maybe 0.25dB here or there. That kind of thing -- yes, i'll do. However, making substantive changes outside of apparent artists intent, then I ask for feedback. Also, the stereo image needs to be modified before decoding -- how much of a change i needed? Answer: I listen for expander surging, and when I find it, I fix the problem. Sometimes poeple want a tone control -- the listener has a tone control. if I make a change JUST FOR MY TASTE, then I always get in trouble... I don't make changes for taste, I do changes for bug correction. John
  14. Here are the new ones coming off the presses right now... These are MUCH more sane... Thanks, Alex!!! The final version will be just a BIT more tame yet, but these are close. THIS IS HARD STUFF TO DO, it isn't like a tone control at all!!! First one is 'Dreams', Second one is 'Second Hand News' https://www.dropbox.com/s/wgtjk5xbzpmihox/DreamsNew.flac?dl=0 https://www.dropbox.com/s/6o1nwz6ydb2945a/SecondHandNewsNew.flac?dl=0
  15. Will help modern digital recordings ONLY if they are feralA. I have found some material that is, but it shouldn't happen all that often. Most new material that might be feralA is when/if it was created back in the past. John
  • Create New...