FLAC files from 16/44 CD (WAV) files are smaller than MP3s.

[esldude]

You say your DSD is 352 khz.  Is that DXD?  DSD typically goes 64, 128, 256, and 512.

[esldude]

Another thought, Yuri is a member here, and writes software to do this for Mac and Windows.  Maybe even his software you have.  Perhaps he could help.  

 

audiventory

 

Is the user name for Yuri here.  

[esldude]

http://www.sonore.us/dsd2flac.html

 

Here is a free conversion for Mac or Linux.  I have used it myself.  I didn't suggest it before because it only does DSD64 and 128.  Converts to Flac up to 352 khz rates.  Rather simple, but works.  You could try that and see how the results turn out versus what you have done already.   

[esldude]

1 minute ago, dalethorn said:

 

https://www.dropbox.com/sh/st6goiq1q0mjmsz/AACgOP4f6dQLD4m0KLWlb8yJa?dl=0

 

Here's a DSD and the JRiver-made 16/44 WAV.

 

 

I get an error saying the file can't be found on your link. 

[esldude]

I converted to 24/88 with the Sonore software.  Works fine.  

 

The FLAC from the conversion was 39 meg.  Opening it and saving as a wav resulted in 78 meg.  So 50% compression which is about right for FLAC.  The 16/44 wav downloaded from dropbox was 26 meg which also is in line with a 78 meg wav in 24/88 form.  Considering the parameters of DSD64 being close to 96/24 in size the 100 meg size of it seems about okay too. 

 

Glad to hear a recording using the Isomike setup anyway.  Thanks. 

[esldude]

Okay, I took your 44 wav file.  Compressed it to 24 bit Flac and about 45% size reduction.  When I compressed it to 16 bit I see what you mean.  The file size was 6 meg or so.  I also have never seen one compress that well.  I have played with FLAC before and signals with extremely low signal levels and curtailed highs will compress less than 50% though I haven't seen one compress this much.

[esldude]

Okay, I took the 88/24 wav file from the Sonore conversion.  Opened it in Audacity, exported it as a 44/16 level 5 Flac.  The result was slightly larger than compressing your 44 wav file at 7 meg (your file compressed to 6 meg).  Still a smaller size than is the norm. 

 

I think mansr is correct, it simply is a file that compresses well due to no high frequencies and low noise levels in the recording. 

 

BTW, if you use the Sonore converter.  It has an option to adjust the output level.  Default is +4 db.  I find this to come close to matching PCM files of the same material.  This was 4db higher in level than your conversion there at 44 khz. 

[esldude]

http://www.bm.rs/Kimber/Kiber - ISO Mike System - Audiophile Voice.pdf

 

Bonus info.  Here is an article about the Kimber Isomike written by CA's own George Graves. 

[esldude]

2 minutes ago, dalethorn said:

 

It cannot be purely a compression issue, as I've explained many times when converting from MP3 to WAV to FLAC.  The FLAC codec is behaving totally out of character, and only with JRiver's file.

So how do you think I obtained a nearly identical result using all different software?  And kumakuma has also gotten a near identical result with yet another software. 

 

When I compressed to FLAC as a 24 bit file it was only 50% compression. 

 

In any case I'll try Ogg Vorbis and let you know. 

[esldude]

This is interesting.  Beginning with the 88/24 wav I converted to 44 with Ogg Vorbis set on a quality level of 5 out of 10.  The file size was only 2.5 meg.  While I know ogg is more like MP3 that is a lot of compression.  High rate VBR MP3 was 4.4 meg.  I think it is simply a compression issue. 

[esldude]

Okay, I took a download DSD128 file from 2L.  Converted to 176/24 flac.  Opened that and converted to 44/16 wav which was 34 meg.  I took that and saved to FLAC which was 15 meg which is 44%.  I then put a steep roll off at 5 khz on the 44 wav.  Saved as a FLAC which then dropped to 9.5 meg.  28% which isn't so far from the 24 -25% on your file.  This recording had an orchestra in it not just a single piano.  

[esldude]

I just did a Glenn Gould recording from a CD.  Compressed from 16.3 meg to 6.0 meg.  36% which is lower than I usually see.  This recording also drops off about 5 khz.  It is sourced from old reel tapes however and there is low level noise not in the recording you used. 

[esldude]

Okay I just put a rolloff above 5 khz on the Glenn Gould recording to drop out the tape noise.  Wav to FLAC 4.6 meg from a 16.3 original wav file. 26% for that. 

 

I think you have a clean digital recording with nothing above 5 khz and that is the answer. 

[esldude]

Why would I lie to you.  Don't you think this is getting ridiculous now.  Jeez!

 

Here is what I used. 

https://www.amazon.com/J-S-Bach-French-Overture/dp/B0000028NL

 

They list it as a 1995 CD.  The CD says it is a 1972 recording. 

 

 

[esldude]

6 minutes ago, dalethorn said:

 

I bought it from a Amazon provider who promises to deliver Jan 31st.  This should be good, unless someone here closes the topic.

I could have sent you a couple tracks from it.  You seem to wish to have your own copy which is okay.  

[esldude]

https://www.amazon.com/Glenn-Gould-plays-Bach-Variations/dp/B0085MK2M6/ref=sr_1_3?ie=UTF8&qid=1517009355&sr=8-3&keywords=glenn+gould+goldberg+variations

 

Here is one you can get overnight. 

 

I did track 16.  As a wav it is 53.1 meg and as a Flac 16.4 meg.  I didn't do any filtering to it.  I had it ripped from CD as a FLAC on my hard drive.  Opened it in Audacity.  Saved it as a 16 bit wav.  Saved it again as a level 5 16 bit FLAC.  31% size as a FLAC.    I used a 3rd order rolloff at 5 khz.  Saved as a FLAC and this time 23% of the original size. 

[esldude]

3 hours ago, CatManDo said:

Sorry, I didn't read the entire thread, and I'm not an encoding expert. Just an experienced user.

 

Could this have to do with the treatment of mono recordings (the 1955 Goldberg Variations are mono)?

snip...

The Goldberg Variations disc I suggested is a two CD set.  One of the 1955 sessions and one from the 1981 sessions (which were digitally recorded).  I used the stereo track from the latter.

 

Digital tracks will compress better (on a limited range instrument like piano) because there is less high frequency low level noise from tape or noisy analog electronics (like when tubes were used).  Noise doesn't compress much at all.  

 

[esldude]

Or to look at it another way Dale, I took the Goldberg Glenn Gould track.  Put a brickwall filter at 6 khz.  There is nothing other than low level noise above that.  Noise at -100 db.  The RMS level of the track was -27 db before I did it. And the same after the brick wall filter.  All that is up there is noise or harmonics buried in the noise.  One could have recorded this with a 14 khz sampling rate and it may not have been audibly different.  14 khz vs 44 khz is right in that low 30% range it compressed into.  

[esldude]

5 minutes ago, dalethorn said:

 

Not what I said.  The 16/44 WAV already lost info from the original high res source.  Why can't we get a FLAC that's not so extremely compressed, but has more of the original content instead?  The high res distributors issue FLACs without intermediate conversions.

Can you clarify what you mean by this.  

[esldude]

8 minutes ago, dalethorn said:

 

Let's say I have one of those million-dollar file makers and I'm issuing 16/44 FLACs from my high-res masters.  I discover, without going through a WAV step, that my 16/44 FLAC is smaller (or nearly as small) as an MP3.  Why couldn't I generate a larger FLAC that has more info that's still playable on players that don't accept files higher than 16/44 or 16/48?  There must be a way, rather than saying "that isn't the way things are done".

Sounds like MQA.  Or at least what MQA claims to promise.  

[esldude]

4 minutes ago, dalethorn said:

 

For all I remember of the discussions at Stereophile, MQA is intended as a streaming format (or modification).  In a sense, what I'm asking has a similar bent, but I'm not asking for a special codec that requires a special player to play it.  I'm asking about making the FLAC from a higher res master and simply losing less data.

What you are asking for can't be accomplished within the parameters of the format.  If you start with a 192/24 files you have to reduce the bits to 16.  You have to limit bandwidth to that of 44 khz sample rates.  Other than minor variances in the transition band (20khz to 22050 hz) the differences are all in the higher frequency band.  What content that exists below 20 khz is basically being fully captured in the FLAC.  

 

Taking the Gould track for an example.  There just isn't anything above the noise floor above around 6 or 7 khz.  There isn't anything there.  A simple piano recording is not as information rich as more complex music.  Higher rez doesn't capture more because other than noise there isn't anything.  

[esldude]

5 minutes ago, dalethorn said:

 

I'm not so sure.  I think you agree that the 16/44 FLAC is lossy compared to the master, so the question is, what's thrown away?  I appreciate that there's a sample rate that has to be adhered to for the players that can't go above 16/44, but still, someone is making the decision what to discard, and I doubt it is (or has to be) merely cutting the output data by 4 times.  I'm guessing it's more complicated than that, and there are options...

 

What is being lost in this lossy conversion, is any frequencies above 20 khz, and any bit values below 16 bit.  Various downsampling algorithms accomplish this in slightly different manner.  The dither can vary, and how the transition band filtering is handled can vary a little bit.  For the most part none of the below 20 khz info is lost.  There is no direct way to put the lost info in the file. 

 

http://src.infinitewave.ca/

 

Maybe you have seen this.  It shows the results of various down-sampling softwares.  Izotope and Sox are two of the very best methods available.  

[esldude]

Something you can do, is take a 192 file.  Downsample it to 44.  Upsample back to 192.  Subtract one file from the other, and the difference is what was lost in the downsampling.  What you'll find is some very low level noise below 20 khz which is from the dither.  And whatever was above 20 khz.  You'll find you cannot hear any of this unless you amplify it heavily and even then you'll only hear the noise like dither.  

[esldude]

Well, redbook did have a provision for pre-emphasis. Almost like an RIAA curve for LPs.  A gradual slope to +9.5 db by 20 khz.  Then the reverse slope upon playback.  Provided extra resolution of high frequencies beyond 16 bits.  Not done anymore.  Was rare even in the early days.  I think I have 5 CDs with it.  I doubt any modern hardware looks for the pre-emphasis flag or can play it back properly.  

 

Other than that I think you are searching for an explanation that isn't there.  

 

What information can you add?  Essentially nothing below Nyquist is lost and you can't add above nyquist info with it still being a PCM 44.1 format.  You could do stuff like MQA does with the folding and unfolding.  But not something you can stick in a common player and get anything out of it. 

[esldude]

CD rippers will spit out bit perfect rips.  End of the story. 

 

FLAC will do bit perfect lossless compression which is bit perfect reversible.  End of the story. 

 

There is nothing else.  The characteristics of some music is more amenable to higher compression ratios than others.  Piano with quiet noise floors is one example of that. 

 

As far as DSD conversion, if the software you were using doesn't work it is a software problem.  The Sonore software works fine.  I believe in this thread 3 other softwares have been able to convert fine.   What is it you think is going on? 

 

You originally didn't believe piano recordings compressed below 50%.  Now you  have seen that they compress into the 30% range.  What is the problem you are looking to solve?