CD to DSD conversion


Hi all.

So yesterday I decided to use JRiver to rip my CD and I probably have a setting to convert everything to DSD on the rip so it created dsf files.

So I took the time to rip it again in Apple lossless which keeps it native 44/16. There is definitely a difference. I am trying to figure what sounds better and am wrestling with it a bit.

In a blind test I had someone pick the dsf. Their description : sounds more airy … The other seems flatter. Personally - I understand that descr but really am still scratching my head as to what I think is better.

Has anyone else done this and what was your outcome?


How are you playing the files back and into what DAC?


I have a direct stream - playing thru JRiver via USB. As a further note - I am bit perfect and do not upsample using the software. This rip however, was really an accident as I don’t usually rip using JRiver. But I looked at the DS and it showed dsd 64 and my files went to dsf.


So I will try to explain what I think I hear. The dsf files seem to have more sustain / decay on the notes. Voice sounds more open. The 44/16 (alac) file sounds denser… a little more closed in. Someone could define it as ‘solid’ as well. I lean towards the converted dsf file.

The fact I can hear a repeatable difference and pick it out blindly says a lot.

So back to my original question. Has anyone done an a/b on something like this ? Would the DS be ‘happier’ skipping the PCM->DSD conversion (making an assumption). Am I hearing the difference in rippers (iTunes vs Jriver i.e. Does iTunes suck at ripping cd?) or is this conversion to dsf on the rip just better thru the DS?

timm said

Would the DS be ‘happier’ skipping the PCM->DSD conversion (making an assumption).

Rather the opposite - the FPGA architecture is such that every option is running all of the time, then the results of what you want are selected. Think of all if-then-else as running both the “then” arm and the “else” arm and then using a hardware switch to pick the answer you want. This means that you are saving nothing by processing things ahead of time. The DS also uses more accurate upsampling than many algo’s on PC, especially when converting to DSD.

Which algos you use for upsampling and/or PCM -> DSD conversion you use will affect the sound of a DSD vs PCM test. You need to listen to each option you are considering for a week or so and then thy the other - what ever your gut will tell you the answer.

To me the difference between PCM and DSD isn’t anything to due with sound stage, balance, extension etc. (tho DSD does seem more airy to me on the top end.) But rather after listening to DSD, PCM seems a little less solid, a little more ghostly, more like a memory vs live. DSD just sounds a little more real to me.


Yup, the DS is converting everything up, so adding a conversion before is a waste of HD space. ITunes isn’t necessarily the best, and I stopped using Apple lossless a long time ago, as I didn’t see any reason to convert the original files, and at the time, most DACs were native 16/44.1 or 24/96 PCM, and so tended to sound a bit better ‘native’. Haven’t really done any comparisons since, though, as HDs just kept getting cheaper.

One of the things I found to be important in ripping was using a decent external disc drive (not necessarily expensive) rather than the built-in drive in a laptop or a Mini. I’ve had more than one that you can feel and hear the vibration from the disc spinning. They just tend to be very lightweight. A more stable spinner helps keep the errors down. Then if you’re wired that way and have the time, use Max or some other bit checker.


I was using the drive that came with my dedicated music player laptop. It took a crap, and with Toshiba’s warranty, I would have had to send it in. Someone at the Asylum recommended a Samsung USB DVD drive. I ordered it from Amazon, and when I was setting it up, I found out it was the external version of the built in drive. It lasted less than a month. Amazon took it back. I was looking for a better quality externally powered drive, but at the time, I couldn’t find one, so I bought a Lenovo drive. So far it has worked flawlessly, but is lightweight.

Looking at some music players, one of them offered an external transport, so I dug deeper. It is branded Buffalo, and has external power, the only downside is the price, over a $100. But it will probably be the next drive I buy. If the Lenovo dies, and I have the money available. I just don’t like the idea of a separate drive running on 5v it gets from USB. I was kind of surprised, there aren’t more external drives with their own power, I will check the specs before buy, as I would be very disappointed if I found out it was a 5v transformer.

As to doing format and resolution changes, they should be done during playback. Imagine you rip hundreds of CDs all converted, then you get a new DAC and find it is better with native resolutions. There is another topic where the guy has been experimenting with conversions, you might want to read that. I believe that all your ripping should be bit perfect copies of the disc being ripped.

jeffstarr said I just don’t like the idea of a separate drive running on 5v it gets from USB. . . . I will check the specs before buy, as I would be very disappointed if I found out it was a 5v transformer.
Why? What is wrong with an external drive operating on 5V? The drives are designed to work on this voltage and receive sufficient current.


Well … Bad beef - I know conceptually it would seem a waste to up convert - save to disc and play. Yet results are indeed ‘different’ as compared with letting the DS do the work on the fly. Whether one likes one result over the other would be subjective.

Ted - indeed the difference I hear seems to correspond to your ‘DSD is more airy’ comment. I mean the DS does different things - branches to different code via your if- then-else example based on type of file received ?

I guess I can try ripping a cd with JRiver native and up converting to hear the difference. This would take the iTunes rip out of the equation.


FWIW - I’ve had horrible experience with external drives powered from an AC converter (more than 1/2 outright died but ALL corrupted data after a time.)

On the other hand I’ve got 11 USB powered drives (some of the smaller (in space) ones USB 2 and the others USB 3.) All of them My Passport or My Passport Ultra. Some are older than the DS. They’ve performed flawlessly (but see below.)

The key thing if the drive is plugged into a USB hub is that the USB hub be powered externally and that the power supply supplies all of the current that the (relevant) USB spec specifies for all of the downstream USB connections running at full current: 0.5A for each USB 2 output and 0.9A for each USB 3 output. Even so with Windows 10 and 5 or more USB drives on one hub I need to unplug all of the drives from the USB hub before I boot the PC and plug them back in one at a time after the PC is booted otherwise trying to power up all of the drives at once is too much for my 10 output USB 3 hub (and it has a big external power supply.)

timm said

Ted - indeed the difference I hear seems to correspond to your ‘DSD is more airy’ comment. I mean the DS does different things - branches to different code via your if- then-else example based on type of file received ?

No, like I said all of the code is running all of the time, no branches - only the answers that are relevant are looked at. FPGAs don't execute one instruction after another, they do everything all of the time. If you want you can use part of an FPGA to implement a conventional processor but even then the all instructions are always running in parallel but only the answer from the one that's "currently being executed" is looked at.

For the DS that means that the answers of the upsample to 352.8 or 384kHz block is ignored for DSD inputs. After that all of the processing is identical for all inputs: the 352.8k, 384k, DSD 64 or DSD 128 to 56.448MHz converter, then the volume control/quad rate single bit output sigma delta modulator.

For the technically curious the more advanced the FPGA the more likely that there is internal hardware to save some power in a subset of the FPGA if the outputs of that subset aren’t being used, etc. But I turn all such features off. Unlike conventional programming where we care about optimizing the best case or average case in general the point of real time programming is to make the worst case as close to the average case as possible.


Hmm. Ok. I was just trying to whittle down to the cause of the difference. This is a repeatable blind test result. So being ‘technically curious’ lol , I grasp for reasons. Obviously the ripper used is different. Next is the algorithm used to rip i.e. Up convert to DSD. If I can come up with ‘the reason’ then maybe other leaps of faith could be had which could create a repeatable result. :slight_smile:


timm - if I read the OP right, you’re comparing an Apple lossless rip with a DSD upconvert? The source is Redbook CD, correct? Have you tried ripping it as 16/44.1 AIFF (i.e. No conversions at all) and comparing that file with the airy DSD? And all the resulting files are being played back on a computer via JRiver?

Not saying there will be a difference, but it would be interesting to see.


This will be my next step BB. I will do it… For the sake of science. :).


Sometime ago I was curious and started to rip from redbook then convert the resulting file to DSD64 using the Tascam Hi-Res Editor. Gave up almost right away as the DSD files sounded noticeably worse than the real native DSD ones, no contest really.


Do you mean ‘real native PCM ones’? I get why a native dsd file would sound better than one upsampled to DSD


Ok. Because I’m kind of a geek… I put 3 files in JRiver. iTunes alac /flac from JRiver / and up converted DSD from JRiver. I have a DS direct to amp with Maggie 20.7s. My opinion - the flac and iTunes file were fairly indistinguishable. The up converted DSD file seemed to continually have a bit more high end and I thought sounded better. Something I could blind test and pick. I did it once - as did my wife. This seemed to give the music more jump in my opinion. All of the rips were to my c drive

For whatever reason - it just seems to have a bit of what I think is space, and high end. More difficult to tell on bass but I do think it is better. When I say more decay …sustain… It could be just the higher end doing that for me on say a cymbal or an (is it E string) acoustic guitar.

Cd used : Some Devil. Dave Matthews

Tracks focused on: Gravedigger(acoustic)/Grey Blue Eyes/Oh/ An’ Another Thing

this is actually a pretty decent recording of a rock type of cd. Lots of acoustic guitar, percussive instruments, cello.

the Maggie ribbon tweeter is pretty revealing.

I guess what I will quantify it as is that the cd flac/alac rips just sound a little muted in frequency in comparison. Is this a huge diff - probably not really … But if I can pick it out - to me it’s relative.


timm - still no unconverted file? I would not expect the two lossless files to sound different.

I know most feel that “lossless” should/must/has to be lossless, and maybe it usually is - “bits are bits”, right? - but sonically it seems to depend on how good/transparent the encoding and decoding is. (Is it HW or SW, does it cause the computer to be working while it’s playing, etc.). The data may all be there, but it also may be accompanied by other new noise, etc.

Then there’s the question of how JRiver/your rig is handling/processing the different types of files…

Not trying to be a PITA. Just trying to sort out what might be happening, rather than arriving at the conclusion, “JRiver DSD upsamples sound better than letting the DS do it” (if that’s what you’re suggesting). I am in no way saying that’s not what’s happening here, one just has to be careful what conclusion you draw or infer from it. There is a fair amount of HD space and hassle at stake here, potentially.


Beef, if I am understanding correctly, the OP is doing the conversion to DSD during the ripping process. This is the first time I have heard of anyone doing it this way. It is not unusual for people to do a conversion during playback. This is something new, to me, I wouldn’t do it, preferring to rip bit perfect, and let my DAC do the converting. The Lindemann does not do DSD, so I do have JRiver converting those DSD files to 24/176, and they sound very good. But I certainly wouldn’t convert them to PCM, replacing the original format. If I ever go from being broke to buying a new DAC, it will be DSD capable.

The OP may be on to something, it is certainly different.

Elk, I should have said an external transport using the same current, as the 5v it is pulling from USB. Those USB transports, the three I have experience with are all quite flimsy. Then it has to run a motor to spin the disc, operate the laser, and control it’s sled. I just looked at the Buffalo Melco recommends, it is pricey compared to the other USB drives, but it is what I would buy, if I could. I looked, it runs on 12v, see link.

Ted, you are either lucky, or with your skills, you have optimised those USB drives. When the internal drive quit working, I checked the drivers, and in properties, it claimed it was working, but I forget, I had problems with one doing burns, and one quit reading, so ripping was out. If I have to replace the Lenovo, I found a site that tested a bunch, and recommends the LG.

Here is the Buffalo I would like to have:


It is interesting how our experiences differ. I, like Ted, have had nothing but good experience with USB powered drives, both HDD and optical. I have never bothered to try external powered drive as they add additional complexity I have never felt I needed.