Why you want upsampling and upconversion in a server and why it sounds way better!
One of Roon's and Jrivers best advantages is the ability to digitally reprocess your data.
With an appropriate server you can take for example a 16 bit 44k Tidal stream and reprocess that data stream to DSD or to reprocess to 24bit 192k, 384k, even up to 768k.
In our tests almost accross the board most listeners perfer DSD or high res PCM data streams.
The counter arguement is that if you start with 16 bit 44k sample you can't do any better as that is the source.
Going through Facebook we got a post that showed the Lumiere brothers first film shot in 1894 with a steam locomotive.
The second video showed the same Lumiere video reprocessed using AI to create a 4k video stream out of the same video footage.
The first video is the native source the second shows the reprocessed video the differences aren't subtitle even if you reprocess the native video you are able to extract a signifigantly higher quality image.
We have been saying this for years even if you don't start out with a true high res source through the magic of computer reprocessing you can create a pseudo high res data packet that still sound far better than ifs original form.
Watch the video and see what you think. Sure it is always best to start with a true high resolution image but in the case of not having a true high resolution music fille the ability to use digital reprocessing can create a signal that will sound far better providing that your dac and server are up to the task.
Any novice photographer could have told you that. The two forms of media that you are referring to have very little in common when it comes to neural networks, auditory subtleties and individual hearing variances.
There has never been an objective study that reliably proves that the different upsampled songs can be identified at a percentage better than random guessing, even by “golden ear professionals”.
What you have delivered is just an opinion, which everyone is entitled to. 😎🎼
Mind you, Sony claims they do the same with some of the MP3 restoration work they are doing, but up and oversampling algorithms do not. At worst, they extrapolate a line. At best they use a curve fitting algorithm. Far from AI based, but interesting to think about for the future.
In my mind, DACs used to perform better with high rez data, and that's changed recently. The quality of performance between Redbook and high resolution has vastly improved.
The video/film analogy is a stretch IMO. However, my impression is that the general consensus is that upsampling Redbook CD's usually does result in audible improvements, and I agree. I've been doing it for years, though in hardware rather than software.
CD's clearly improve when upsampled through my old-but-beautiful Assemblage D2D-1 upsampler/dejitterer. It feeds a Sonic Frontiers SFD-2 MK-III via I2S-e interface. This system only goes to 24b/96khz, but so what? How many of us can really hear the difference between 24/96 ans 24/192 or whatever? [And please, nobody whine about the old kit or "digital has improved dramatically," etc etc unless you've actually heard it. It's bold, but I'll say it - at the very high end this stuff hasn't actually been sonically improved on. But all that is a topic for another thread.] What sonic improvements? Very subtle but also very real. More space and depth in the soundstage. Clearer placement of instruments, less "smeared." Just a little clearer, better defined and simply more real.
Most significant to me, is that the same improvements occur when playing upsampled 16/44, "CD quality" material from Quobuz and Tidal through the BlueSound Node 2i. In this case it's the s/pdif output of the Node 2i to the D2D-1, thereby bypassing the DAC in the Node 2i. The D2D-1/SFD-2 sounds >>vastly<< better than the Node 2i's internal DAC.
Interestingly, playing Tidal MDA material through the D2D-1/SFD-2 (which wouldn't be unfolded or decoded or whatever, I guess) sounds better than when using the Node 2i's internal DAC with full MDA unfolding. Admittedly, this probably isn't a fair comparison.
Sorry guys you just don''t watch the video. It is clear that the 4k Lumiere video is much more watchable then the original.
There is no contest between the two film examples.
For most people upconverted or transcoded audio mirrors the same experience with video. Farjouda video processors were a huge improvment over early high quality video display technology in the 90's and it was clear to see that a reprocessed video image was clearly better than the original one that was not.
In our tests with many high end dacs the upconverted data nearly always sounds far better than the original.
I have the digitally massaged, cleaned up, improved and higher fidelity LP of old songs from the soundtrack of “Once Upon a Time in Hollywood”. At first listen it’s impressive as heck and I’m sure it would get a giant thumbs up across the board too. The songs are so much clearer and fidelity is massively improved. Can’t listen through even one side. The tunes are great, it doesn’t connect. Some things are better left alone.
I have experimented with upsampling. I had the Chord DAVE with Blu 2 and then M Scaler. Upsampling to 768k initially sounded impressive but over time it came to sound unnatural and fatiguing. The sound was more detailed, but it was artificial detail - lean, skeletal and unnatural. It was also bright and tonally bleached. It’s the sort of thing that impresses in a short demo but ultimately sounds unnatural and artificial.
I sold the DAVE and upscaler and now have the Lumin X1, which can stream 768k files. So i tested the 768 upsampling in Roon. It had exactly the same effect - more apparent detail, but lean, bright, bleached and fatiguing.
Playing files at their native resolution - or with minimal upsampling - sounds best to me - fuller, warmer and more natural. All the detail is there, it just isn’t emphasised in a grossly unnatural manner.
The analogy with video images is a false one - the two work very differently.
There are no free lunches in audio or anything else. Adding digital information to a file may produce some benefits but as always there must be a cost somewhere. In my view that cost - the unnatural, artificial, fatiguing sound - far outweighs the miniscule benefits.
wlutke -- I only half-agree with you on the question of whether older/imperfect recordings should be massaged. On the one hand, I appreciate any increase in fidelity. After all, isn't that why we buy all those expensive cables? On the other hand, there's no denying the sheer charm and mellifluousness of, how shall we say, vintage recordings. And who knows what Al Jolson or Arturo Toscanini might have thought?
I use Roon to upsample my Lumin T2 playback to DSD 512. I absolutely love the sound, better sound staging side to side, back to back plus a more natural presentation of details. I don't get fatigued or hear a sound that is bleached or unnatural sounding. Another of example of different ears in different rooms with different gear. I agree completely with Audiotroy's post.
I use Roon to upsample my Lumin T2 playback to DSD 512. I'm using a SmallGreen Computer as my Roon server. I absolutely love the sound; better sound staging side to side, back to back plus a more natural presentation of details. I don't get fatigued or hear a sound that is bleached or unnatural sounding. Another of example of different ears in different rooms with different gear. I agree completely with Audiotroy's post.
2L Recordings has a bunch of content, hyper well recorded in many formats - a free workout and critical listener development tool. But most A/B here on audiogon at least don’t own and use an spl meter, so louder wins... also, Native DSD ( not converted PCM ) has a nice sampler download.
I’ve dabbled in upsampling with Roon and my Lumin D1 with mixed results. Some music sounded a little better upsampled to 96khz, but not really worth getting excited about. No real difference beyond that in frequency. However having Roon send the stream as DSD was worth the trouble. I’ve only recently tried this and am still exploring the differences. So far the ideas of more spacious and defined soundstage and more musically enjoyable seem to be evident. I like it so far.
While I enjoy Dave and Troy’s posts and listening opinions, on this one I’m also going to have to disagree.
As a video pro I can say that analog video cannot compare in any way to digital video. However we all know that analog audio still can be be very high quality. And for instruments like drums, analog is the preferred way of recording on most high end rock albums - the analog is then brought into Protools. No one is using analog video anymore.
As far as upsampling - I’ve found that the native sample rate sounds best, when done in the computer.
Oversampling in the DAC ( or like PS audio DS that uprez to DSD) seems to be more successful.
However I do remember on the old PS Audio PerfectWave, when they rolled out the nativeX mode option, it sounded better than oversampling, which many of us were using before the NativeX proved superior.
I have to agree with many of you. At first listen, upsampling seems to make things sound much better, smoother, more analog and natural. But continued critical listening reveals issues of unnatural, fatiguing sound that eventually has me running back to a native bitrate. Ideally, I've found that a 24/96 native recording played back without up- or down-sampling seems to be the sweet spot. But I listen to thousands of well-recorded 16/24 CD's ripped to my server that sound so great I never want for more. A well-recorded Redbook CD is a wonder to hear.
That being said, in many cases streaming music sources can sound better upsasmpled. And mp3 files, at any bitrate, sound much better upsampled to at least 24/96. With good upsampling mp3 can almost sound listenable.
My impressions of this are not limited by equipment, or source materials. I've tried this experiment on the highest of high-end DAC's, servers, CD transports, etc. in world-class reference-quality systems, with every bitrate and quality of recording, and the issues always eventually reveal themselves.
On my setup I listen to everything through my server at native bitrate, with NO digital manipulation (delays, EQ, balance controls, etc.). Over thousands of hours of listening I've come to the conclusion that I want to hear what was recorded, as it was recorded, warts and all. I don't want a polished turd ...I'd rather know it's a turd and learn to live with that. In my collection I have many great perfomances that were recorded horribly. I'm OK with that, because my great performances that were recorded well sound like heaven.
The more resolution your room and system provide, the easier it is to hear these differences. In an acoustically isolated high-end system the fatigue of upsampling is impossible to ignore. At least that's been my experience...
Ditto from me too, up/over/sideways sampling all BS, just another way to sell more dacs to the gullible.. Should just have a good listen to Redbook 16/44 PCM converted through a "good" discrete R2R Multibit nos dac
So I just tried upsampling to DSD 128. It takes up about 5-10x more CPU in the Roon server, I think this is as high as I can go with my pokey A10 CPU.
My DAC (Mytek Brooklyn) has issues switching sample rates. Sometimes it glitches, which is annoying when going from my personal library to MQA/Tidal to internet radio. It doesn't happen every time, but it is annoying, so sample rate converting to a common standard removes this issue.
I've done most of my upsampling to 768k using PCM. I have just been experimenting with DSD 512 upsampling on my Lumin X1.
Pretty much the same results. DSD always tends to sound a little smoother anyway, but in a way I find artificial. The upsampled sound produced a slight background glare that I did not care for, and certainly did not sound better than playing it at native PCM resolution.
OK, having just done some informal comparisons I am now absolutely ready to render the definitive verdict on the subject. For me. :)
I went from straight PCM streaming to DSD 128. I agree with @rossb - Mostly what I heard was in the range of cymbals. Little that I heard was all that in terms of information, it felt like the top 2 octaves were turned up, and far too much at the cymbal range.
It did not seem natural, but I can see why in a quick listen it may be impressive. I did not hear better sound stage for instance, or deeper fuller bass. What I heard was in the range of selecting different filter settings in a DAC.
In the end, I’m not keeping this setting active. If I want the same effect I'll turn the treble up.
I remember you saying previously that your Brooklyn gave issues occasionally when switching between differing resolution rates.
Not sure if that is a trait of the Brooklyn though as I cannot say I ever had it happen on my Brooklyn?
I’ve had at least one other person say it happens to them as well, but did not delve further. I’m listening exclusively to PCM via USB. This may not happen over coax or optical. It is not a big enough deal for me to ask it to be fixed, I just hope some day to be able to afford to trade it in for a Manhattan. :)
Also, I have kind of suspicions it has to do with how hot the DAC gets, so I try to leave it elevated and away from the amp.
I know this happens when I hear music, but it sound like beez are buzzing around in the speakers. I have to switch back to whatever mode it is stuck in, and then switch again.
Also, it has done this with a variety of Linux servers and software (Squeezelite, Roon) acting as sources. Would be interesting if this was for instance a Linux driver only sort of issue.
maybe it is a function of the USB input from n the earlier Brooklyn DAC?
Maybe. :) It definitely feels like a software/timing glitch to me. Coax/optical have no handshake. The DAC must continuously monitor the incoming signal and sync to it, unlike USB where there is definitely a handshake and after that the DAC pulls each frame.
I’ve been round the mulberry bush several times with upsampling to my PS Audio DSD DAC. I know why Roon’s upsampling can be appealing. Depending on how it’s done, I can hear the apparent increase in space and detail around the notes. However, this seems to be dependent upon which phase filter is being used.
The phase filters as applied in Roon definitely add their own flavor to the music. To me, the minimum phase filter smooths things out, kills dynamics a bit, and seems to smear the music ever so slightly.
Conversely, while the linear phase filter does raise the spectre of pre-ringing, it also gives that apparent increase in space and detail you referenced above; almost like a Photoshop sharpening, or contrast filter. But, this seems to come at the expense of removing a bit of sweetness from the highs (I’m sure there are other differences). To me, it leaves the sound a little more harsh. On short listens, I know exactly why people might prefer the sound.
For me, all roads have lead back to native frequencies. I have yet to find any DSP combination in Roon that is more appealing than a native file, in the long run. Although short-term changes seem to make improvements, after the novelty wears off, I have without exception felt the exchange was a bit of a Faustian bargain (trading the music’s soul for cheap tricks).
Have you considered whether your preference for upsampling is because of the phase filter being used? Something of a “sound effect”?
My Roon core is currently on a MBP, and I recognize how busy and electronically noisy those can be.
I’m one of those that want to like upsampling. I like to think that I’ve tried every combination of upsampling in Roon; though likely, I haven’t. I think you’re right to point out the different architectures in DAC’s . The native upsampling in the PS Audio Directstream DAC could absolutely contribute to perceptual differences when compared to upsampling in Roon. Its very possible that DAC’s from one manufacturer perform better with upsampling, than DAC’s from another manufacturer.
I have upsampled using the smooth, minimum phase filter, then 7th order CLANS to DSD, off and on for a couple of months. I think I’ve tried most of the other settings as well. I’m not saying I’m writing off upsampling as a legitimate method of increasing SQ, I just haven’t found the settings that would make me prefer it over native streaming on my particular DAC. I’d be interested to hear what you like best. I’m not above trying something new.