Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[dalethorn]

On the Mac, I converted some 352 khz DSDs to 44 khz "Uncompressed WAV", then converted those to FLAC at several levels (1 to 6, tried 'em all), and the FLAC files are as small as 320 kbps MP3s.  I played those files in Foobar2000 on a PC, and also did FLAC conversions of the WAV tracks on the PC, and no matter what, Foobar says they're ranging from 290 kbps to 336 kbps, which corresponds to their filesize.

When I rip a CD to 44 khz WAV tracks on the PC and then convert the WAVs to FLAC (level 5), the FLAC filesizes range from 60 to 70 percent of the WAV filesizes.  But the conversions from DSD are about 23 percent the size of the WAV files, no matter what the FLAC level.

I have no idea what's going on (my conversion screen turns off everything except the 44 khz destination sample rate), but it sure looks like MC is not letting me get a true lossless WAV file from the DSD files.

[Awesome Donkey]

It's worth pointing out that converting DSD to PCM (WAV, FLAC, etc.) is a lossy conversion, even if you convert it to a lossless format like WAV or FLAC.

Also is MC set to converting (dithering) to a 16-bit, 24-bit or 32-bit bit-depth?

[dalethorn]

It's worth pointing out that converting DSD to PCM (WAV, FLAC, etc.) is a lossy conversion, even if you convert it to a lossless format like WAV or FLAC. Also is MC set to converting (dithering) to a 16-bit, 24-bit or 32-bit bit-depth?

As I pointed out, MC is making 16/44 WAV files with 1/3 the content of a 16/44 CD rip that I do myself.  Something is very wrong.

EDIT: I changed the Subject line to reflect that the "44 khz" WAV file contents are only 290 KBPS, i.e. about the same as an MP3.

[dalethorn]

The reason I got into these conversions is because a Stereophile writer told me "conversion is easy" using MC, so that I could feel at ease buying DSD albums that are not available in other formats.  If it turns out that MC makes 16/44 WAV files whose actual content is about the same as an MP3, then in my view that's a form of DRM being passed on from the DSD files.

So can anyone confirm that MC will create CD-equivalent WAV files from DSD files?  If MC can do that, then when I convert those WAVs to FLAC in Foobar2000, the compression at level 5 will be about the same as I get from my CD rips.  If not, then someone here should know what's happening.

Edit: When I tried creating 24/88 WAV files on MC, just trying to play those files corrupted my Foobar install.  Foobar has no problem playing all of the hires files I buy from HDTracks and other sites.

[JimH]

A 16/44 WAV file isn't at all the same quality as an MP3 file.

It is the same as the track on the CD.

[dalethorn]

A 16/44 WAV file isn't at all the same quality as an MP3 file. It is the same as the track on the CD.

The actual issue is that MC seems to be making a WAV file containing only a fraction of the data it should have.  As I explained, when I rip a CD and convert to FLAC with Foobar2000, the FLAC size is about 63 percent of the WAV size.  When I convert a DSD to a 16/44 WAV in MC and then convert that WAV to FLAC in Foobar2000, the FLAC is 23 percent of the WAV size.

So MC's "lossless" WAV file is really MP3 data in a WAV container.

23 percent versus 63 percent.  I've done a good bit of research on WAV to Level 5 FLAC conversions, and the 23 percent result is way out of bounds.

[JimH]

The actual issue is that MC seems to be making a WAV file containing only a fraction of the data it should have. 

I don't believe that's the case.

[dalethorn]

I don't believe that's the case.

But that is the factual result that I'm getting.

[Matt]

I converted a 1x DSD file to FLAC and the resulting FLAC was huge and 352,800 Hz.

I wonder why you might be seeing something different?

Double-check that you don't have any DSP on for the conversion.

[Awesome Donkey]

I think dalethorn is purposely converting DSD files (and dithering/resampling) to 16-bit/44.1 kHz WAV files and then converting those 16-bit/44.1 kHz WAV files to FLAC in MC. To be honest, I've never tried that.

Just a random thought, do you have the SoX resample option enabled? If not, you might want to try it (if so, try disabling it).

[RD James]

There seems to be a number of potential issues here.
 
First of all: filesize and bitrate is no indicator of audio quality. Many SACD/DSD sources do not contain high resolution audio - they're just converted from a CD master.
Secondly: you cannot compare SACD conversions to a CD rip. It's entirely possible that they are sourced from different masters.
Thirdly: why go from DSD to 352.8kHz WAV to 44.1kHz FLAC? Media Center can do the conversion directly.
 
I don't have access to a Mac right now, but on Windows at least, Media Center does the conversion correctly.
Make sure that Media Center is properly configured for it though:

• Tools > Options > Audio > Settings > Use SoX for Resampling
• Tools > Options > Audio > Advanced > Dither Mode: TPDF Dither.
• Enable DSP, and make sure that the only option you have selected is Output Format in DSP Studio (set to stereo 44.1kHz).
• Set the bit-depth to 16-bit.
• Set the FLAC Encoder to Verify Encoding.

Now Media Center should be configured to convert directly from DSD to 16/44 FLACs.

One tool you can use to check the encoded files is Spek.
If the FLAC file is lossless, it will either have data extending to 22.05kHz or natural looking peaks. If it's sourced from an MP3, it will have a hard cut-off at 16kHz.

• Lossless WAV
  
• Compressed MP3

I had to go through three different SACD albums before I found a clear example like that though - the other two discs didn't have frequency information that even went up to 22kHz.
Posting the result here may be useful.

[dalethorn]

There seems to be a number of potential issues here.
 
First of all: filesize and bitrate is no indicator of audio quality. Many SACD/DSD sources do not contain high resolution audio - they're just converted from a CD master.
Secondly: you cannot compare SACD conversions to a CD rip. It's entirely possible that they are sourced from different masters.
Thirdly: why go from DSD to 352.8kHz WAV to 44.1kHz FLAC? Media Center can do the conversion directly.
 
I don't have access to a Mac right now, but on Windows at least, Media Center does the conversion correctly.
Make sure that Media Center is properly configured for it though:

• Tools > Options > Audio > Settings > Use SoX for Resampling
• Tools > Options > Audio > Advanced > Dither Mode: TPDF Dither.
• Enable DSP, and make sure that the only option you have selected is Output Format in DSP Studio (set to stereo 44.1kHz).
• Set the bit-depth to 16-bit.
• Set the FLAC Encoder to Verify Encoding.

Now Media Center should be configured to convert directly from DSD to 16/44 FLACs.

One tool you can use to check the encoded files is Spek.
If the FLAC file is lossless, it will either have data extending to 22.05kHz or natural looking peaks. If it's sourced from an MP3, it will have a hard cut-off at 16kHz.

• Lossless WAV
  
• Compressed MP3

I had to go through three different SACD albums before I found a clear example like that though - the other two discs didn't have frequency information that even went up to 22kHz.
Posting the result here may be useful.

I tried your settings, adding the SoX resampling and TPDF dithering, and the improvement was from 23 percent of WAV size to 23.8 percent.  The WAV size is 538302828 and the FLAC size is 128535037. Playing the FLACs in Foobar shows 302 to 346 kbps.  Basically, MP3 data in a WAV container.

BTW, there is no way on Earth to convert a normal WAV file to FLAC at 23 percent of the WAV filesize, using a reliable converter.  There are lots of research online that demonstrate those limits, even at the FLAC Level extremes.  I don't know if there's a bug here, or something else.  But my results are consistent.

The reason I'm doing 16/44 currently is because the higher res conversions actually corrupt my Foobar installation.

[dalethorn]

I think dalethorn is purposely converting DSD files (and dithering/resampling) to 16-bit/44.1 kHz WAV files and then converting those 16-bit/44.1 kHz WAV files to FLAC in MC. To be honest, I've never tried that.  Just a random thought, do you have the SoX resample option enabled? If not, you might want to try it (if so, try disabling it).

With SoX selected and the TPDF dithering enabled, the difference in FLAC size increased from 23 percent WAV size to 23.8 percent.

[RD James]

I tried your settings, adding the SoX resampling and TPDF dithering, and the improvement was from 23 percent of WAV size to 23.8 percent.  The WAV size is 538302828 and the FLAC size is 128535037. Playing the FLACs in Foobar shows 302 to 346 kbps.  Basically, MP3 data in a WAV container.

That does sound low, but as I said before, FLAC bitrate should not be taken an indicator of audio quality. It's more about audio complexity.
Just because the resulting file compresses well does not mean that it's being sourced from MP3 data.
In fact, if you convert an MP3 to a FLAC file, you will likely end up with a much higher bitrate and larger filesize.
Solo Piano music often compresses very well with FLAC for example.

BTW, there is no way on Earth to convert a normal WAV file to FLAC at 23 percent of the WAV filesize, using a reliable converter.  There are lots of research online that demonstrate those limits, even at the FLAC Level extremes.  I don't know if there's a bug here, or something else.  But my results are consistent.

Is it possible for you to upload one of the source tracks somewhere?

[dalethorn]

That does sound low, but as I said before, FLAC bitrate should not be taken an indicator of audio quality. It's more about audio complexity.
Just because the resulting file compresses well does not mean that it's being sourced from MP3 data.
In fact, if you convert an MP3 to a FLAC file, you will likely end up with a much higher bitrate and larger filesize.
Solo Piano music often compresses very well with FLAC for example.Is it possible for you to upload one of the source tracks somewhere?

I think I've covered the complexity issue well, given that I have *lots* of classical piano just like this DSD album.  My tastes are rather narrow, so if it doesn't fit my complexity biases, I discard it.

I'm uploading the smaller DSD file and the WAV I created in MC to my Dropbox account.  It may take a dozen more minutes.  I'm not sure I remember Dropbox procedure - do I send you a link of some kind?

[dalethorn]

Dropbox files are ready - how to make them visible to you?

[Hendrik]

I believe you can just generate a public link for any file you uploaded to dropbox, so it can be shared.

[dalethorn]

I believe you can just generate a public link for any file you uploaded to dropbox, so it can be shared.

Try this:

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

EDIT: I sent someone a PM (now deleted) asking for email address for sharing Dropbox, but maybe this link will work instead.

[dalethorn]

Is it possible for you to upload one of the source tracks somewhere?

Try this:

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

EDIT: I sent someone a PM (now deleted) asking for email address for sharing Dropbox, but maybe this link will work instead.

[RD James]

Got it, thanks.
It's exactly what I was describing before: low complexity solo piano music, without much in the way of high-frequency content, and no ultrasonic audio content.
It compresses well because there isn't much there.

• If I disable the low-pass filter - which you should never do during playback - and output a 24-bit 352.8kHz WAV file to display the raw audio, you can see that there is a gap between the highest frequency content in the track, and the DSD noise at ~30kHz (DSD is a very noisy format).
  
• If I re-enable the low-pass filter to eliminate the DSD noise, you can see that a 352.8kHz file is looking very empty.
  
• And when you convert that to a 16-bit 44.1kHz WAV file, it basically cuts out the empty space, since the file has no ultrasonic content at all. Even then, the track is looking quite empty, because it's a solo piano piece.

There's nothing wrong with your converted files, and it's not being encoded as an MP3. It just happens to compress very well.

[dalethorn]

Got it, thanks.
It's exactly what I was describing before: low complexity solo piano music, without much in the way of high-frequency content, and no ultrasonic audio content.
It compresses well because there isn't much there.

• If I disable the low-pass filter - which you should never do during playback - and output a 24-bit 352.8kHz WAV file to display the raw audio, you can see that there is a gap between the highest frequency content in the track, and the DSD noise at ~30kHz (DSD is a very noisy format).
  
• If I re-enable the low-pass filter to eliminate the DSD noise, you can see that a 352.8kHz file is looking very empty.
  
• And when you convert that to a 16-bit 44.1kHz WAV file, it basically cuts out the empty space, since the file has no ultrasonic content at all. Even then, the track is looking quite empty, because it's a solo piano piece.

There's nothing wrong with your converted files, and it's not being encoded as an MP3. It just happens to compress very well.

I can't argue with your numbers since I don't have your tools, but I can refute your conclusions on a "reality" check.  Name one CD of piano music that you think will compress well, and I'll rip WAV tracks and convert those to FLAC, and the maximum compression will be above 50 percent.  Cannot happen any other way.

That BTW is the same experience reported on a plethora of Internet sites.  Now, just to be clear, I don't know what kind of content is in the DSD files, if that makes a difference.  However, the maker of these DSD files are going to reissue the recording as 24/96 FLAC and lower res PCM files.  When they do that, I will obtain the 16/44 WAV version and convert that to FLAC.  Then we will see.

In the meantime, feel free to point out even one CD that will compress to 23 percent of WAV size and still be lossless.

[dalethorn]

BTW, what the above guy is reporting as "high compression due to low complexity" is exactly what lossy compression does.  Look it up.  FLAC compression cannot compress this piano music to 23 percent for the simple reason that it's not extremely low complexity, and it has to be lossless.

[Matt]

BTW, what the above guy is reporting as "high compression due to low complexity" is exactly what lossy compression does.  Look it up.  FLAC compression cannot compress this piano music to 23 percent for the simple reason that it's not extremely low complexity, and it has to be lossless.

I use APE and I have piano albums compressed to 23 percent or even less.

[dalethorn]

I use APE and I have piano albums compressed to 23 percent or even less.

Great - if one of those that compress to <23 percent is on an available CD, tell me so I can buy it and test it.  Thanks!!

BTW, I sent the "low complexity" message to Ray Kimber, who recorded these DSDs - I'll forward his response when he replies.

[tyler69]

BTW, what the above guy is reporting as "high compression due to low complexity" is exactly what lossy compression does.  Look it up.  FLAC compression cannot compress this piano music to 23 percent for the simple reason that it's not extremely low complexity, and it has to be lossless.

The guy above spent probably some time into educating you and others in this forum on why the filesize of your compressed file is that low. He did it in a understandable way even for laymans like me. You should try to understand what he actually wrote and the technical implications thereof.

[dalethorn]

The guy above spent probably some time into educating you and others in this forum on why the filesize of your compressed file is that low. He did it in a understandable way even for laymans like me. You should try to understand what he actually wrote and the technical implications thereof.

So "science" trumps experience?  No.  Firstly, let's get that CD and learn something there.  Secondly, let's hear from Ray the recording engineer.  And if more reading and "understanding" is needed in the meantime, there are a lot of audio gurus on the Internet who can describe their experiences.  Remember that I have lots of CDs and downloads, so I'm quite capable of testing FLAC compression.

[tyler69]

Do you have similar experiences with digital music that has a signal that goes above 10kHz?

[mojave]

I converted the DSD to 16 bit 44kHz wav with both JRiver and ConverteR. Similar file size.

I converted the 16 bit 44 kHz wav to FLAC with JRiver, Audacity and ConverteR. Similar file size.

I converted the 16 bit 44 kHz wav to APE and ALAC with JRiver. Similar file size.

You can try as many programs as you like, but the results will be the that the solo piano piece you uploaded converted to any lossless compression format will be about 23% the size of the wav file.

BTW, 18 years ago Matt wrote the Monkey's Audio (ape) algorithm for lossless data compression. His wife also considers him a concert pianist. He understands exactly how solo piano music can be compressed to 23%.