|
Post by johndyson on Mar 23, 2020 4:40:26 GMT 10
You know that the results are getting better, but not quite the pinnacle of perfection. Every darned time I get ready to put the DHNRDS DA/FA decoder set aside to work on the C4 decoder, then I find another way of improving the DHNRDS.
It seems like the results are as perfect as I know how to make them, the lo+behold, I find another improvement. It seems like I come up with new ideas when writing these notes to people.
Let me tell you what is happening right now... A few days ago, I had decided to increase the resolution of some of the math (in critical places) from single precision to double precision. Most of the very critical filters are already double precision, but I found another place where some improvement is possible, so I did it. Guess what? When reviewing the code, and increasing the precision of some of the operations, I realized that there is a place where more filtering can be applied, with zero loss of quality. This filtering is intended to avoid creating some of the undesired sidebands during gain control -- and the results are even more silent and precise sounding. The clearest place to find an improvvent are the 'vocal enhancement' used back in the '70s, where they would use a single or double echo chorus effect... The amount of detail, when improved also changes that sound of 'brightness' and splits it into the individual chorus entities.
So, the bad news is that there will be new decodes appearing again. The good news is that there will be new decodes appareing again!!! It is hard to imagine that the results are better, but they really are. The actual first phase of the decoding (producing the .wav files from the decoding operation itself) is now complete, so now simply need to produce the compressed flac files. I'll be making the decoding results available some time tomorrow (about 28Hrs from posting time.) The reason for the delay is that there are all kinds of other decodes happening now also... Too many things going on, but I will be making the results available to Alex, and then he can pass them on. I might simply post one album at a time so that Alex can pass on the raw .wav files. My .wave files have a BeXT extension in them, and have a serial number that I'll provide. Poor man's MQA (YUCK!!!!) :-).
John
|
|
sandyk (RIP Alex, 1939 - 2021)
Global Moderator
Posts: 226
About Me: Retired ex Principal Telecommunications Technical Officer with 43 years at Telstra (Australia)
I am also a Moderator in Hi Fi Critic Forum
Electronics hobbyist for >65 years with DIY projects including Loudspeakers, Stereo FM tuner, S/W Regen Receiver, Superhet AM ,
Synchrodyne PLL AM tuner (Phase Lock Loop),Stereo Tape Deck, Amplifiers including I.C. types, Class A, Class AB 100W/Ch. (ETI5000) 240W/Ch. Mosfet (AEM6000) ,several DACs , numerous PSUs including VERY low noise (<4uV) types etc.for myself and friends
Audio Industry Affiliation: NIL
|
Post by sandyk (RIP Alex, 1939 - 2021) on Mar 23, 2020 9:03:36 GMT 10
You know that the results are getting better, but not quite the pinnacle of perfection. Every darned time I get ready to put the DHNRDS DA/FA decoder set aside to work on the C4 decoder, then I find another way of improving the DHNRDS. It seems like the results are as perfect as I know how to make them, the lo+behold, I find another improvement. It seems like I come up with new ideas when writing these notes to people. Let me tell you what is happening right now... A few days ago, I had decided to increase the resolution of some of the math (in critical places) from single precision to double precision. Most of the very critical filters are already double precision, but I found another place where some improvement is possible, so I did it. Guess what? When reviewing the code, and increasing the precision of some of the operations, I realized that there is a place where more filtering can be applied, with zero loss of quality. This filtering is intended to avoid creating some of the undesired sidebands during gain control -- and the results are even more silent and precise sounding. The clearest place to find an improvvent are the 'vocal enhancement' used back in the '70s, where they would use a single or double echo chorus effect... The amount of detail, when improved also changes that sound of 'brightness' and splits it into the individual chorus entities. So, the bad news is that there will be new decodes appearing again. The good news is that there will be new decodes appareing again!!! It is hard to imagine that the results are better, but they really are. The actual first phase of the decoding (producing the .wav files from the decoding operation itself) is now complete, so now simply need to produce the compressed flac files. I'll be making the decoding results available some time tomorrow (about 28Hrs from posting time.) The reason for the delay is that there are all kinds of other decodes happening now also... Too many things going on, but I will be making the results available to Alex, and then he can pass them on. I might simply post one album at a time so that Alex can pass on the raw .wav files. My .wave files have a BeXT extension in them, and have a serial number that I'll provide.Poor man's MQA (YUCK!!!!) :-). John John Is that the Header information that Sound Forge 9 found, and removed when saving those 32/88.2 files again ? As you are well aware I do prefer .wav over .flac for SQ reasons which appears to be mainly due to how electrically quiet the PC/Server is.
Kind Regards Alex
|
|
|
Post by johndyson on Mar 23, 2020 19:20:12 GMT 10
I just tried to verify my ability to provide the full-monty wav versions and I cannot do it because of space. Because of space reasonns, it looks like I just might be able to do the 48k/24 flac versions. The files are HUGE -- the FP format is much bigger than even the uncompressed 24 bit wav versions. THe only reason why I put the label infromation on the files is so that they can be referred to. I have a tendancy to improve things over time, and it is really nice to be able to look directly at the file contents. This is a simialr reason why pros add the metadata to the files also. My flac files also keep metadata so that people really know which is the latest version (I usually update the serial numbers and decoder versions in the files when there is a major improvement.) I am trying to improve my processes so that no-one (including me) gets the versions confused. When there might be a complaint or comment about quality on a given version, it is nice to be able to look directly at the files and KNOW which version the file is (and all of the internal date information also.) Since there is ZERO information (other than dither) in my files above the frequency response of the original CDs, then 48k can give everything. The only advantage of 88.2k is the dither density can be less, so the noise floor is slightly lower -- there is zero additional information. So, theoretically an 88.2k/16 file is slightly better than a 48k/16file, but not by much. The 48k/24 files can represent MUCH MORE information than my decodes can produce. The best amount of effective resolution that can come out of the decodes would be on the order of 18bits. (16 bits plus about 2 bits for the DolbyA NR.) Everything beyond the 18bits of effective detail is noise.
John
|
|
|
Post by johndyson on Mar 23, 2020 21:03:44 GMT 10
One more comment -- I'll describe the data items I plan to maintain in any .wav or .flac file that I produce. There will be a serial number -- not really a serial number per-se, but is actually just a sequence number/reference ID for my builds. There'll also be a raw version ID for the decoder. I haven't standardized yet, but the info is in the files. Eventually, I plan to keep the same serial number for each decoding version, and then that serial number will be automatically transferred to the flac files. This is not a 'security' thing as much as a reference ID. I might change the nomenclature to 'reference ID' so it is less confusing. No matter what, for my own sanity I need to keep some relationship between the decoder version, build date and decoding attempt -- just so I know what is going on in the future. Already, since the very first day of development, I have practically every version of the decoder ever built -- well, when being actively developed at least 3 snapshots per day. Also, at a moments notice, I can actually run an ancient version of the decoder... The difficult part is remembering the command set, since it has changed drastically since the beginning.
The Supertramp stuff has a first version of the scheme, but when I started upgrading my ABBA build procedures, I realized that maybe the Supertramp scheme isn't quite correct. The scheme will be reconciled before the next decoder release on Tuesday along with the upcoming demos.
I think that the build IDs will have a common sequence ID for every build that I do. The decoder has been modified to be able to write its version ID into a command script, but it almost always produces the BEXT data for serious professional use. I have found that sometimes when I lose track of a file, just being able to reference what version of decoder being used can be nice.
John
|
|
sandyk (RIP Alex, 1939 - 2021)
Global Moderator
Posts: 226
About Me: Retired ex Principal Telecommunications Technical Officer with 43 years at Telstra (Australia)
I am also a Moderator in Hi Fi Critic Forum
Electronics hobbyist for >65 years with DIY projects including Loudspeakers, Stereo FM tuner, S/W Regen Receiver, Superhet AM ,
Synchrodyne PLL AM tuner (Phase Lock Loop),Stereo Tape Deck, Amplifiers including I.C. types, Class A, Class AB 100W/Ch. (ETI5000) 240W/Ch. Mosfet (AEM6000) ,several DACs , numerous PSUs including VERY low noise (<4uV) types etc.for myself and friends
Audio Industry Affiliation: NIL
|
Post by sandyk (RIP Alex, 1939 - 2021) on May 19, 2020 8:56:40 GMT 10
If any member would like to hear a corrected sample track from John Dyson of Dire Straits-Love Over Gold, please send me a PM. It should sound slightly better again than the original highly sought after Vertigo 088-2 CD from W.Germany.
Alex
|
|
sandyk (RIP Alex, 1939 - 2021)
Global Moderator
Posts: 226
About Me: Retired ex Principal Telecommunications Technical Officer with 43 years at Telstra (Australia)
I am also a Moderator in Hi Fi Critic Forum
Electronics hobbyist for >65 years with DIY projects including Loudspeakers, Stereo FM tuner, S/W Regen Receiver, Superhet AM ,
Synchrodyne PLL AM tuner (Phase Lock Loop),Stereo Tape Deck, Amplifiers including I.C. types, Class A, Class AB 100W/Ch. (ETI5000) 240W/Ch. Mosfet (AEM6000) ,several DACs , numerous PSUs including VERY low noise (<4uV) types etc.for myself and friends
Audio Industry Affiliation: NIL
|
Post by sandyk (RIP Alex, 1939 - 2021) on May 19, 2020 8:58:19 GMT 10
If any member would like to hear a corrected sample track from John Dyson of Dire Straits-Love Over Gold, please send me a PM. It should sound slightly better again than the original highly sought after Vertigo 088-2 CD from W.Germany.
Alex
|
|