Computer Audio Asylum

Music servers and other computer based digital audio technologies.

Return to Computer Audio Asylum

Message Sort: Post Order or Asylum Reverse Threaded

dsd to flac

176.57.33.44
Posted on August 1, 2015 at 22:43:50
maxim
Audiophile

Posts: 458
Joined: May 22, 2000
Contributor
  Since: September 24, 2013
 when converting with dbpoweramp, what sample rate/bit depth should I use? any other recommended settings?

I can play dsd in one system, but not in the other one ...

thanks.

maxim

 

Hide full thread outline!
    ...
RE: dsd to flac, posted on August 2, 2015 at 04:17:34
bullethead
 I would recommend 24bit 192khz FLAC. It should sound pretty good. I would also use the lowest compression level, less to process at the hardware level.

 

RE: dsd to flac, posted on August 2, 2015 at 05:40:42
RadioWonder
Audiophile

Posts: 1236
Location: Arizona
Joined: March 16, 2003
 I prefer "0" compression using dbpoweramp Music Converter... It just sounds better...

 

96 or 192, posted on August 2, 2015 at 06:23:11
maxim
Audiophile

Posts: 458
Joined: May 22, 2000
Contributor
  Since: September 24, 2013
 Dbpoweramp recommends 96 khz to cut off supersonik noise ...

I have no idea if it makes any sense

 

RE: 96 or 192, posted on August 2, 2015 at 06:31:40
bullethead
 supersonic noise is the whole reason for DSD and high res flac. It has an effect, I would do 24 bit 192.

 

RE: 96 or 192, posted on August 2, 2015 at 06:45:11
Roseval
Audiophile

Posts: 1846
Joined: March 31, 2008
 Inherent to DSD is the quantization noise.
In case of DSD 64 it starts at 22 kHz and in case of DSD 128 at 44
The DSD standard says to apply low pass filtering (50 kHz if I remember correctly) to avoid your gear being bombarded with high frequency artifacts inherent to the DSD to analog conversion.
Hence half fs= 96/2 = 48 kHz is not a bad value.
It by and large depends on the filter used in the conversion.
You might try a tool like MusicScope to compare the DSD with the PCM conversion.

The Well Tempered Computer

 

24/176.4 or 24/88.2, posted on August 2, 2015 at 06:51:32
Kal Rubinson
Reviewer

Posts: 12436
Location: New York
Joined: June 5, 2002
 for integer math.

 

RE: 24/176.4 or 24/88.2, posted on August 2, 2015 at 10:17:26
maxim
Audiophile

Posts: 458
Joined: May 22, 2000
Contributor
  Since: September 24, 2013
 Kal,

Could you please explain this a little bit.

Thanks.

Maxim

 

RE: 24/176.4 or 24/88.2, posted on August 2, 2015 at 10:57:01
Kal Rubinson
Reviewer

Posts: 12436
Location: New York
Joined: June 5, 2002
 The sample rate of SACD is 2.8224 MHz which is evenly divisible by 44.1K, 88.2, 176.4, 352.8, etc. The conversion math is simple. Converting to rates that are not even divisors is messy.

I tend to use 176.4 so that I can apply Dirac room correction on the fly.

 

RE: 24/176.4 or 24/88.2, posted on August 2, 2015 at 12:49:03
maxim
Audiophile

Posts: 458
Joined: May 22, 2000
Contributor
  Since: September 24, 2013
 I see, makes sense.

And in your experience no need to worry about ultrasonic noise?

 

RE: 24/176.4 or 24/88.2, posted on August 2, 2015 at 14:22:01
Kal Rubinson
Reviewer

Posts: 12436
Location: New York
Joined: June 5, 2002
 If I was an equipment designer, it would be of concern to me to insure that my equipment is stable regardless of the signal bandwidth and I would take into account today's signal formats.

As a user, no.

 

Neither, play the DSD., posted on August 2, 2015 at 18:29:34
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 Ultrasonic noise is a problem when downsampling DSD to a lower sampling rate. If the filters do not prevent all aliasing, then the ultrasonic noise will fold down into the "audible band". On the other hand, if the filters do eliminate aliasing then the result may be filter artifacts and unintended loss of sound quality compared to the original DSD. The solution to this quandary is to avoid downsampling DSD to lower sampling rates. Get a DAC that will play these files in their original format. If this is not possible, then downsample them to the highest available sampling rate.

Given that you have to downsample them until you get a suitable DAC, then by all means use the highest sampling rate available. It will do the least damage. However, even going to 352.8 is not going to be transparent compared to the original DSD.




Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

RE: Neither, play the DSD., posted on August 3, 2015 at 15:24:54
Kal Rubinson
Reviewer

Posts: 12436
Location: New York
Joined: June 5, 2002
 Well, I can play the DSD but the sound is superior if converted to 24/176.4 and processed through DiracLive. Still, a comparison the DSD with the PCM, on an equal basis, accords with yours.

 

RE: Neither, play the DSD., posted on August 3, 2015 at 19:35:28
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 If you can get a set of convolution impulses out of Dirac live, then you can possibly put the DSD through HQPlayer and get the same room correction without loss of any resolution. This is what I do with 2 channel DSD recordings, but my processor is not fast enough to handle multiple channels or to upsample the result to DSD128.

I use a parametric equalizer, no phase response, for my room EQ. I put an impulse through the VSP equalizer with the settings that I chose and then passed the resulting two second impulse file to HQPlayer. It can prcoess it without requiring any downsampling. at least assuming the correction is just in the bass. I don't believe this will work for multichannels without a real fast (e.g. gaming) CPU. My dual core 3.2 GHz i5 650 can barely handle two channels of DSD at DSD64 in and out.

Definitely the room correction is more than worth the degradation in sound quality with a down conversion to 176.4 if you can't stay at 2.8 MHz sampling rate.


Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

RE: 24/176.4 or 24/88.2, posted on August 4, 2015 at 18:06:20
dtc
Audiophile

Posts: 102
Location: Boston Area
Joined: March 31, 2006
 This is the common logic when converting PCM to different sample rate PCM - that you use an integer multiple. I am no expert, but I have read multiple places that the DSD to PCM conversion does a lot of calculations that loss any advantages of doing integer conversions for DSD to PCM. Are you sure the integer multiple argument really applies to DSD to PCM? Can you point to any discussion of the conversion process to support the integer idea? Thanks.

 

RE: 24/176.4 or 24/88.2, posted on August 5, 2015 at 08:59:17
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 The general consensus is that converting DSD to PCM will work better when done to even multiples. This was certainly the case with some software and hardware converters a few years back when comparing DSD to 88.2/24 vs. 96/24. It could well make less difference at 176.4/24 or 192/24 or with different software.

For example, Bruce Brown, the mastering engineer who did many of the early HDtracks.com SACD rips and PCM conversions, reached the conclusion that 88.2 was better than 96. I believe this was discussed in this forum quite a few years back.





Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

RE: 24/176.4 or 24/88.2, posted on August 5, 2015 at 11:56:23
dtc
Audiophile

Posts: 102
Location: Boston Area
Joined: March 31, 2006
 I understand the common consensus, but I still wonder if a lot of that is simply a hold over from PCM conversions. I have seen several discussions that indicate that the DSD to PCM conversion process so scrambles the original data with floating point calculations that the integer multiplier idea just no longer applies. With PCM, you can anchor some of the points with the original data. I am just not sure what the parallel is in DSD to PCM conversion. I am still looking for some references to how the calculations are actual done to try to understand the issue.

 

Converting DSD to PCM, posted on August 5, 2015 at 12:43:49
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 The situation with PCM conversion has substantially changed over the years due to improvements in software and faster processors. This is particularly true where upsampling is concerned. Previously many sample rate programs had gross errors when upsampling 44/24 to 96/24 that were absent in converting 44/24 to 88/24. This is not true with the newer conversion software such as the iZotope 64 bit SRC.

To convert from DSD to PCM you do the following:

1. You treat the 1 bit samples as +1 and -1 floating point numbers
2. You low pass filter these to ensure that there is no aliasing, i.e. no energy remaining above 1/2 the output sample rate. You also have to roll off the high frequencies starting at 25 kHz if your low pass filter hasn't already done this.
3. You high pass filter to remove (or minimize) a DC offset problem arising because the DSD samples can not directly represent 0.
4. You scale the resulting floating point output to numbers in the range of -1.0 and +1.0 and then convert these to 24 bit fixed point (with dither).

There are some other minor issues that have to be addressed such as muting and unmuting at the ends of a file, etc... The choice of filters is where all the art is.

You can use either floating point calculations or integer calculations, provided that you have at least 48 bits of accuracy. The choice depends on the available hardware and its performance. If you do not have enough processing power then you will have to take shortcuts and then you may definitely need to keep things synchronized, i.e. it may not work well at all downconverting to non-integer ratios, unless clever programming is involved.

Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

88.2/24 and don't look back..., posted on August 5, 2015 at 13:31:33
PaulN
Audiophile

Posts: 1412
Joined: January 13, 2000
 For a number of reasons: first, all you are encoding above 88.2kHz is the rising out-of-band noise of DSD. Second, the files are more compact, easier of network resources and the sound quality is arguably better because you are not sending all the noise to the DAC to deal with. Finally, as Kal says, the math is integer so there is no interpolation. I encode all of my DSD64 to 88.2/24 and DSD128 to 176.4/24.

 

RE: Converting DSD to PCM, posted on August 5, 2015 at 14:34:14
dtc
Audiophile

Posts: 102
Location: Boston Area
Joined: March 31, 2006
 Please bare with me - how do you change the +1, -1 numbers into PCM samples? Do you simply take a running sum of the +1,-1 numbers and take each 16th one for 176 KHz and each 14.7th one for 192 KHz? Or do you average each group of 16 or 14.7? Or is there some other process involved?

 

RE: Converting DSD to PCM, posted on August 5, 2015 at 15:45:29
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 They +1.0 and -1.0 are floating point numbers. Consider them to be, for all practical purposes, real numbers. When you are done with the filtering you will have a bunch of real numbers in the range of -1.0 to +1.0 (after possibly having to multiply by a scale). Then you have to convert these numbers into fixed point if you are going to send them to a fixed point DAC. (Or the operating system may do this for you.) If you want 16 bit numbers you would multiply by 16383 to get a number in the range of -16383 to +16383 and then convert this sign magnitude representation to a 2's complement number, which is what DACs are going to accept. The numbers for 24 bit would be -8388607 to +8388607. This is all basic computer arithmetic.

The filtering does the necessary averaging. (Averaging is just a form of filtering.) After you are done the filtering, etc., to downsample to 176.4 you take every 16th filtered result and discard the rest. This process is called decimation. Obviously, if you are clever you don't have to calculate the numbers that you are going to throw out.



Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

RE: Neither, play the DSD., posted on August 5, 2015 at 17:27:26
Mr_Steady
Audiophile

Posts: 2042
Location: North Florida
Joined: August 19, 2014
 >"Definitely the room correction is more than worth the degradation in sound quality with a down conversion"

Would it be possible with a program like HQ Player to take a hi res PCM file, alter it with the EQ settings you want, then convert it to DSD, and save the file? You could then load that file onto you local pocket server. Neither the server nor the DAC would have to do any processing, just play the file.

How about, DSD > PCM > EQ > DSD?



-------------------------------------------------------
Big speakers and little amps blew my mind!

 

RE: Neither, play the DSD., posted on August 5, 2015 at 17:36:51
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 At least with my setup there isn't an easy way to save the output of HQPlayer. I'm sure it would be possible to cobble this up with the appropriate ASIO based shims, but I haven't found the need.

It is possible with readily available software to generate the room correction that I used. Actually, I started with measured frequency response and then used a parametric equalizer (VST plug in) to generate the required response. When I got this to measure and sound like I wanted, then I captured the impulse response and used this with HQPlayer. I can use VST plugins with other player software and get the same effect, but only for PCM as I the only PCM to DSD conversion that I have is that built into HQPlayer.

Incidentally, with my Mytek, I don't find it worth while to upsample PCM to DSD64 using HQPlayer. I prefer the original PCM or upsampling to 176.4 or 192. However, upsampling to DSD128 is another matter, probably because this is a sweat spot for this DAC.




Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

RE: Neither, play the DSD., posted on August 5, 2015 at 17:45:45
Sprezza Tura
Audiophile

Posts: 4585
Location: New York City
Joined: August 24, 2012
 DSD128 is the sweet spot for DSD, not for your DAC.

 

RE: Converting DSD to PCM, posted on August 5, 2015 at 18:00:56
dtc
Audiophile

Posts: 102
Location: Boston Area
Joined: March 31, 2006
 Thanks Tony.

I understand the computer math. Lots of PDP 8, PDP 11, VAX assembler, database, device driver, real time, etc. programming, computer science professor type.

So, I guess the question becomes does the filter math maintain the integer multiple integrity that makes 176KHz more desirable than 192 KHz conversion?

 

RE: Neither, play the DSD., posted on August 5, 2015 at 18:08:54
Mr_Steady
Audiophile

Posts: 2042
Location: North Florida
Joined: August 19, 2014
 Thanks.

I just use REW and an analog graphic equalizer. Crude but effective. I have to EQ mids, and highs some.

I'm sure this is a built in sweet spot, but I do like the conversion to DSD on my ES. However I do not like the upsampling to DSD 128, whether it's native DSD or PCM. It makes the transient attacks sound too sharp. Sure I would like to hear PS Audio 10x DSD.



------------------------------------------------------------------
Big speakers and little amps blew my mind!

 

RE: Neither, play the DSD., posted on August 5, 2015 at 18:12:18
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 DSD128 is as far as my DAC goes. I haven't tried DSD256, so I have no way of commenting definitively on your suggestion, but you could well be right.

I listened to Bruce Brown's DSD Battle Royale and found that the Grimm ADC (DSD64 only) was not really bettered by any of the newer DSD128 DACs. So this makes me think that, at least when it comes to ADCs today, that DSD128 may be at, near, or even slightly beyond the point of diminishing returns for purist production processes, e.g. analog to DSD128 to analog with no intermediate conversions or processing except "razor blade cuts". DSD256 may well be necessary for more traditional studio production processes with multiple generations of DSD required for post-production mixdown or sweetening, or worse (e.g. pitch correction as needed for singers that can't sing). I expect it is much easier to make a DAC perform well at DSD256 than an ADC, based on what the circuitry has to do.




Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

RE: Neither, play the DSD., posted on August 5, 2015 at 18:21:51
Sprezza Tura
Audiophile

Posts: 4585
Location: New York City
Joined: August 24, 2012
 Based on my extensive testing between the recording in DSD64/128/256, DSD128 lives up to the promise of DSD.

As a matter of fact, DSD128 make DSD64 sound quite ordinary.

My tests by the way, were based on about 30 recordings done by a friend of mine who works for a small boutique label in Europe. He sent me raw dsf and dff files. These were half native recordings in each DSD flavor, and the other half analog tapes, again archived in each DSD flavor, along with some 24/192 PCM dubs of some of the same tapes.

DSD256, from my tests, offered no audible benefit to my ears over 128.

The files sizes were massive too boot.

All of the above are from my personal experience, and I don't claim this to be some universal truth, but suffice to say, everyone I had listened to the files agreed with me as did the engineer.

 

RE: Converting DSD to PCM, posted on August 5, 2015 at 18:27:34
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 The problem is that if you don't do integer downsampling then you can't do a simple decimation. Instead, you have to interpolate points in between where you put the +- 1's. This can be done, but you have to have a separate bank of filter coefficients for each intermediate point of interpolation. This may or may not be a problem in practice, e.g. it will depend on cache sizes. If you don't use really accurate arithmetic and really accurate filter coefficients, then there will be time varying errors caused by the interpolation process and these will be at the beat frequency, potentially in or near the audible range. I suspect that if done properly (with sufficient horsepower) then the more complex filtering for 192 could well sound as good as 176.4, but I can't see what the point of doing this would be. Going the other way, I have no problem using HQPlayer to upsample 96 or 192 to DSD128. This sounds better then upsampling 96 to 192 or playing 192 directly on my Mytek DAC.

If you want to learn how sample rate converters work then there are text books on digital signal processing that explain this, but you will need college level calculus to understand them (and a lot of time and patience).






Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

RE: Converting DSD to PCM, posted on August 5, 2015 at 18:56:08
dtc
Audiophile

Posts: 102
Location: Boston Area
Joined: March 31, 2006
 I understand the decimation issue. I am asking about the filtering process. Does it maintain the integer relationship?

I have graduate level mathematics, Ph. D. in physical chemistry, post doc in physics. I am just looking for some relatively short, but detailed explanation of the filtering process. I just cannot find it. If you have a source, short of a book, I would be happy to work through it. Other comments I have read indicate that that process does not maintain an integer relationship. That is what I am trying to understand.

 

RE: Converting DSD to PCM, posted on August 5, 2015 at 19:45:20
Tony Lauck
Audiophile

Posts: 13629
Location: Vermont
Joined: November 12, 2007
 You can do the filtering at the original sample rate to provide the needed suppression of aliases. This is just a matter of making sure that the low pass cut off is below FS/2 of the output sample rate. If the output sampling rate is an integer divisor then no interpolation is needed to get the final points, just discarding and reindexing. However, if it is not an integer multiple then you have to do interpolation between the points you do have (at the original sampling rate, but now filtered). You will find two articles in the linked web page that may be useful.

Converting from 44100 to 48000 can be done by converting to the least common multiple which is 7056000 and then downsampling by an integer multiple. This approach will work so long as the two sampling rates are related by a rational number.

44100 * 160 = 7056000
48000 * 147 = 7056000

Another fact is that the samples will line up for all of the common sampling rates every 1/75 of a second, which corresponds to the frame rate on the CD audio format (CDA). Because CD tracks have to be an exact multiple of frames many tracks are integer multiples of 1/75 of a second, with the number of samples depending on the frame rate. This means it is possible to place a set of markers in a continuous sound file at frame boundaries and then resample it to a variety of sampling rates that will be compatible. (Necessary when mastering CD audio of live concerts where playback is to be gapless.)

1 frame: 588 samples at 44100, 640 samples at 48000, 37632 samples at 282400.






Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

 

Page processed in 0.032 seconds.

[ Top Top of Page ] [ Contact Us ] [ Support/Wish List ] [ Copyright © 2024, Audio Asylum, all rights reserved ]
[ General ] [ Speakers ] [ Tubes ] [ Vinyl ] [ Digital ] [ Hi-Rez ] [ Video Asylum ] [ Cables ] [ Tweaks/DIY ] [ Music ] [ Films ]