Playing DSD files

[sngreen]

Hello,
I have Matrix Element X and it does not play DSD files without rather loud and mostly annoying clicks in between the titles. In Asset UPnP there is an option to play DSF files as WAV, which means it is transcoding from DSD into PCM format. I have been noticing the quality of sound is audibly better when playing FLAC files as WAV (likely because the task of conversion is done on the server and not on the final rendering device, thus keeping it less "busy" - I would not know) but is it to be expected for the quality of sound to drop when transcoding from DSD to WAV in return? I mean, DSD is the best sounding format (when mastered from the closest source, of course), but what can be done when the rendering device just does not handle it as well as was intended (and goes without saying advertised).

 

[AnalogSteph]

I have been noticing the quality of sound is audibly better when playing FLAC files as WAV (likely because the task of conversion is done on the server and not on the final rendering device, thus keeping it less "busy" - I would not know)

There has to be another explanation. FLAC was literally designed to be as lightweight as possible on the decoding side, with speeds of 425x-450x on a 2.8 GHz Core 2 Duo T9600. Rockbox on ARM decodes it around 3 times as fast as MP3, at around 8 MHz for realtime.

Transcoding might be discarding / ignoring ReplayGain information, which would result in a level difference. The would be quite noticeable.
If the playback setup involves a ground loop in unbalanced cabling, all bets are off.

but is it to be expected for the quality of sound to drop when transcoding from DSD to WAV in return?

No, at least I certainly wouldn't. Synchronous resampling at high quality is not a major challenge these days, and target sample rates tend to be so high as to be covering frequencies far beyond the audible range, so everything removed would only be shaped noise anyway. You may actually want to not go overboard with target sample rate for this particular reason... ultrasonic noise is at best irrelevant and at worst detrimental to following stages, and it's not helping lossless compression one bit.

Here's a thread on comparable DSD and PCM formats. TL;DR: 24/88 is probably fine.

 

[sngreen]

There has to be another explanation. FLAC was literally designed to be as lightweight as possible on the decoding side, with speeds of 425x-450x on a 2.8 GHz Core 2 Duo T9600. Rockbox on ARM decodes it around 3 times as fast as MP3, at around 8 MHz for realtime.

Transcoding might be discarding / ignoring ReplayGain information, which would result in a level difference. The would be quite noticeable.
If the playback setup involves a ground loop in unbalanced cabling, all bets are off.

No, at least I certainly wouldn't. Synchronous resampling at high quality is not a major challenge these days, and target sample rates tend to be so high as to be covering frequencies far beyond the audible range, so everything removed would only be shaped noise anyway. You may actually want to not go overboard with target sample rate for this particular reason... ultrasonic noise is at best irrelevant and at worst detrimental to following stages, and it's not helping lossless compression one bit.

Here's a thread on comparable DSD and PCM formats. TL;DR: 24/88 is probably fine.

Converting DSD (*.dsf files) to FLAC does reduce gain and changes the overall perception when played along with it. Can't say it sounds muffed, but somewhat less dynamic for sure. Streaming FLAC as WAV does also sound better to my ears, don't know what it is but I would say somewhat cleaner of sorts. But this does not surprise me, as different servers apparently do handle data differently (somewhat), which in turn affects how this data is rendered on the receiving end; as MinimServer vs. Asset UpNP for example (I prefer later). But does packing *dsf files into the wavpacks change how those files sound as well? - that was my question. So far I have discovered that wavpack tag (APEv2) is not very compatible with many servers, those files just do not show listed. But say if they did, would they also sound different?

 

[ZolaIII]

DSD64=PCM 176400 Hz @16 bit (bit perfect)=24 bit DoP (8 bit flags +16 bit data). Wavepack doesn't change DSD bit stream just aply the data compression upon it, still it can do better if you convert it to PCM WAV and then do compression.
ESS DAC's do have a artifacts (click - pop) when switching lines (PCM - DSD and vice versa) but only when it switches in between. If you like DSD do opposite, convert output to DSD128 native but for storage convert the DSD's you possess to PCM WAV (flac).

 

[sngreen]

DSD64=PCM 176400 Hz @16 bit (bit perfect)=24 bit DoP (8 bit flags +16 bit data). Wavepack doesn't change DSD bit stream just aply the data compression upon it, still it can do better if you convert it to PCM WAV and then do compression.
ESS DAC's do have a artifacts (click - pop) when switching lines (PCM - DSD and vice versa) but only when it switches in between. If you like DSD do opposite, convert output to DSD128 native but for storage convert the DSD's you possess to PCM WAV (flac).

But converting to PCM WAV is the same as converting to FLAC, not? Which is exactly what I do not want to do.

Re. APEv2 tag, I read somewhere that those files will show listed as long as the size of the tag does not pass certain limit. Which in turn means I would have to change every single Folder.jpg file to a smaller size, and then retag the files, again - I don't think I will do it either. But I wonder why I do not hear clicks when playing *.dsf files through other streamers, but only through the Element X. Even the Denon home receiver plays them without clicks. That is kind of a bummer, really.

 

[ZolaIII]

ESS has a complex DSD design including physical separate and better authended line.
I stated already that conversion is (exact) bit perfect. Meaning you don't really need to keep it as DSD bit stream on your storage device. Keeping the same in Flac saves you more than half storage space (2x compression). Flac is WAV with data compression, WavPack daos compression part better (better than both Flac and Apple losseles) but it's not widely present. You can and should cut obove 35 KHz on DSD output. WavPack is very interesting in lossy hybrid mode with corestion (restore) graph. Where you can get 50~33% (depending on source complexity) smaller output file for HiFi 88200~96000 Hz 24 bit (or above) than Flac while not being able to hear the difrence and having rest stored in correction file. In comparison to rav DSD bit stream that would be 4~5x compression. Combining lossy and correction files you get original bit perfect Wav back. Only problem is that WavPack is not widely spread out (it is part of the FFmpeg and as such easy available on many platforms but that isn't really enough).
Some software (JRiver for instance) can add a bit of silence on the beginning of DSD native reproduction so that such articles won't be audible.

 

[ZolaIII]

Ideal digital signal is 21 bit 44100~48000 Hz PCM for both reproduction and losseles storage purposes (which won't really charge ever). In transport its perfectly OK to use bit rounding (for the sakes of architectural adoption and decoding convince) to 24 or 32 and it doesn't really matter with what you feel remaining bit range as it will stay out of the DAC limitations and human hearing one. As delta-sigma is a single bit stream it's much easier to transport (no need for data bus nor synchronisation [precise oscillators for clocking - jitter]) it will always have it's applications. In case when you want to use DSD as output stream to DAC (what ever your reason might be) you should use DSD 128 (really no need for more [and let it do noise shaping on good enough lv]) and HF cut filter beyond 30~35 KHz (no need to punish neither equipment nor pets).

 

[sngreen]

How do you convert DSF to FLAC?

 

[voodooless]

How do you convert DSF to FLAC?

@mansr made a DSD-enabled version of sox: https://github.com/mansr/sox . Probably it can do DSD to Flac conversion

 

[sngreen]

@mansr made a DSD-enabled version of sox: https://github.com/mansr/sox . Probably it can do DSD to Flac conversion

MacOS, and then I have to build it? I use Windows/Linux.

 

[voodooless]

MacOS, and then I have to build it? I use Windows/Linux.

That's just the readme that github serves. There is a windows version also: https://github.com/mansr/sox/blob/master/README.win32

Should work on all of these platforms.

 

[mansr]

Converting DSD (*.dsf files) to FLAC does reduce gain and changes the overall perception when played along with it. Can't say it sounds muffed, but somewhat less dynamic for sure.

Converting a DSD file to PCM and playing it on the same DAC almost always produces a difference in level unless you're very careful. Which is louder and by how much depends on the conversion program and the DAC. You have to measure the analogue output level to be sure you're making a valid comparison.

 

[raif71]

Ideal digital signal is 21 bit 44100~48000 Hz PCM for both reproduction and losseles storage purposes (which won't really charge ever). In transport its perfectly OK to use bit rounding (for the sakes of architectural adoption and decoding convince) to 24 or 32 and it doesn't really matter with what you feel remaining bit range as it will stay out of the DAC limitations and human hearing one. As delta-sigma is a single bit stream it's much easier to transport (no need for data bus nor synchronisation [precise oscillators for clocking - jitter]) it will always have it's applications. In case when you want to use DSD as output stream to DAC (what ever your reason might be) you should use DSD 128 (really no need for more [and let it do noise shaping on good enough lv]) and HF cut filter beyond 30~35 KHz (no need to punish neither equipment nor pets).

Delta-Sigma DAC can be abbreviated as DSD though I've never seen it used before

 

[freemansteve]

How do you convert DSF to FLAC?

Foobar2000 with a plugin works well.
I have converted a number of ISO and .dsf files (which FB2K plays fine with the SACD plugin) into FLACs, often 24-96 or 24-88, and I cannot tell any difference on playback, especially without my gigahertz ear-trumpets. It does save masses of space on my NAS/RAID though. I have no idea why files need to be saved as WAVs when FLAC exists, but I'm sure someone will know why! Tagging of FLACs inc. album art (MP3tag does it fine) and adding replaygain (FB2K) is useful to me too

 

[howdoesitsound]

sngreen, I have heard clicks sometimes when using my NAS with my Denon 720ae (DLNA). I don’t use that box for DSD files but I have heard that sound at the end of some tracks (all my tracks are FLAC). I‘m pretty sure it is the same tracks and I need to research further and probably re-do those files, I’m guessing.