Deep Dive

    Moving to Computer Audio: A Comprehensive Overview, Part Two

    Issue 145

    In Part One (Issue 144) we looked at the fundamental considerations in building a computer-based high-resolution audio system, creating a music library, DACs and how to best implement them, and audiophile computers and streamers. Part Two concludes the series.

    The Raspberry Pi Phenomenon

    The Raspberry Pi was launched in 2012 by a team from Cambridge University led by Dr. Eben Upton. He envisaged an affordable computer the size of a pack of cards for kids to learn computing with. To make it low-priced (around $50), Upton used a smartphone processor and put in just enough hardware to run Linux. It uses so little power it runs off a smartphone charger. Plug in an existing TV, borrow a keyboard and mouse and one can get online.

    Dr. Upton aimed to sell a few thousand. As of 2021, they have sold 40 million. Raspberry Pis are amazing not only for their low cost but also for their high quality – they are field-proven in extreme environments. Near-space balloons? Check. Volcano monitors? Sure! Underwater drones? Naturally!

    Computer audiophiles figured out that here was an ideal platform on which to build a streamer. The genius of a Raspberry Pi is its GPIO (General Purpose Input/Output) connector, which accepts plug in boards (“hats” in Pi language). An SP/DIF hat turns it into an SP/DIF streamer; connect a 24/192 DAC hat and you have hi-res audio capability. Add a decent linear power supply or two (one for the Pi and one for the hat) and you have a budget high-end component.


    A wide range of hats are available at different price points. Suppliers such as Allo provide step-by-step instructions so you do not have to be a computer engineer. If you can follow a cookbook, you can make your own Pi audio component over a fun weekend, including hardware assembly and downloading and installing the software.

    If all this sounds a bit DIY, it is. Or was, until Bryston announced the BDP-Pi in 2016. The latest Bryston BR-20 is also Pi-based; you can see the unmistakable connectors on the back panel.

    I should warn you – once you have tasted the Raspberry you may become addicted.

    Bryston BR-20 preamplifier, front view.

    Bryston BR-20 preamplifier, rear view.

    Bryston BR-20 preamplifier, front and rear views.


    Endpoint Streamers – a Minimalist Approach

    If a streamer can beat a Windows computer in sound quality by virtue of minimalism, how about further reducing its hardware footprint? Enter “endpoint streamers” such as the Sonore microRendu v1.5.

    Endpoint streamers are designed to further increase sound quality by doing almost nothing, in a good sense. The heavy lifting of running the music library, user interface, Tidal and audio format conversion is done by a separate powerful music server computer somewhere on a user’s LAN (local area network).

    The endpoint streamer receives music data over the LAN from a server, and merely feeds it to a DAC. If you believe in minimalism, this is for you.


    A USB DAC needs a USB endpoint streamer. If you have a high-end DAC from 10 years ago, an SP/DIF endpoint streamer will do nicely. Also, if you happen to be looking for a new DAC, you could choose one with a built-in endpoint streamer and save yourself an extra component.

    Now that you have your endpoint streamer, you need software to make it work and a music server to partner it.


    Roon is a networked audio playback software whose architecture allows the use of endpoint streamers. A Roon Server computer does all the heavy lifting and feeds music data over the LAN to endpoint streamers for audio output.

    One music server can serve many endpoints. If you have several DACs, you can partner each with its own endpoint streamer and connect them all. You can easily switch between DACs for each track depending on your mood. Those running a tube DAC loaded with precious NOS tubes could configure a solid-state DAC for casual listening. I run three DACs in my main system.

    You can fill your house with music by installing, for example, high-end endpoints in the listening room and more economical endpoints in the kitchen, bedroom, even bathroom. They can all be playing different tracks simultaneously. You can control all of them from your iPad, phone, computer or other device and because the system is LAN-based, everything works all the time and there is no need to re-connect everything as you would have to do when using Bluetooth.

    Roon will accommodate a direct connection between a music server and a DAC, but with significant loss in sound quality as previously explained. Roon supports hi-res streaming with Tidal and Qobuz and I wish they would add iTunes in future releases.

    Roon is easy to configure and can grow as your needs change. I have used Roon since 2017, with excellent results.

    Roon Nucleus music server.

    Roon Nucleus music server.


    Roon Server Software

    Roon Server software is best run on a dedicated computer. There is debate on whether differences in the music server make an audible difference. Theoretically, because the music server is separated from the DAC by the endpoint, there should be no difference. Roon themselves say there is no difference. This is another one of those things that the audiophile must decide for himself.

    There are four main versions of Roon Server software: Windows, MacOS, Linux and Roon Optimized Core Kit (“ROCK”).

    The Windows and MacOS versions are self-explanatory.

    Linux versions are available for PC compatibles, QNAP NAS (network attached storage) and Synology NAS. Running Roon Server on a NAS is economical, but a NAS is not a dedicated computer.

    ROCK comprises a Linux operating system optimized by Roon, bundled together with the Roon Server software. Install this on a PC-compatible computer and one is ready to go.

    In my system, the sound quality of ROCK is excellent and far better than Roon Server for Windows (using the same computer hardware). I have also tried Roon Server for Synology NAS but with disappointing results. I offer no scientific explanation for these observations.

    Roon Server Hardware

    ROCK is an excellent choice but it is certified for use only on Intel NUC computers.

    NUCs are little 4-inch square PC compatibles designed for saving space. They can return decent benchmark scores but are not designed to offer consistent speed or sustained performance. This is because their compact laptop-class hardware must throttle down as much and as often as possible to avoid exceeding fairly low power and thermal limits. This is achieved by varying the frequency of the processor clock – lower frequency saves power when the system is idling and higher frequency provides the speed to do work. This is  fine for Excel, but if you believe a music server’s processor should run at a fixed speed, then you might not like NUCs.

    Computer enthusiasts have successfully run ROCK on a wide range of non-NUC, PCs. These are known as “MOCKs” in the Roon community and there is a thread on the Roon forum dedicated to it. You could roll out that spare desktop PC, as long as it’s not more than around eight years old. There is a good chance it will work and should provide good enough sound quality to get a taste of computer audio before deciding to spend significant money. If you’re computer savvy, I recommend you disable the computer’s power-saving features so that the processor runs at a constant speed. Also, putting the computer in another room will eliminate being bothered by fan noise.

    If your interest in computer engineering is not high, an excellent option is Roon’s own Nucleus server. It is based on the Intel NUC board but has been significantly improved to offer high-end sound. The price includes dealer/manufacturer support. Partner it with a decent linear power supply. Audiophiles who require an ultimate Roon server could investigate the likes of LampizatOr or Pachanko Labs. However, the benefit of a super-high-end server on Roon sound quality is not well documented at this time of writing.

    Differences in sound quality from a “decoupled” server are hard to understand scientifically and therefore hard to predict. It will be necessary to audition the setup in your actual system. In my system, ROCK running on a self-built computer handily beats ROCK on an Intel NUC7i5.

    Optimizing a DAC Using Roon’s DSP

    The Roon Server and endpoint streamers are now filling the house with music. It is time to tune up the DACs by using format conversion (“DSP” in Roon language). This is mainly for hi-res audio listening, because 44 kHz and 48 kHz formats should be streamed to the DAC in bit-perfect mode in most cases.

    Look at their data sheets and you will see that even the best ESS, AKM and Burr-Brown and other DAC chips have lower distortion at lower sampling rates. Some audio DACs, due to their implementation of their DAC chips and other internal circuitry, have good sound only at one sampling rate. For example, my Musical Fidelity Tri-Vista 21, with its rudimentary 96 kHz upsampling circuit, comes alive only when you feed it exactly 96 kHz (to intentionally bypass its internal upsampling).

    In my system, on 24/352 material, the superb-measuring RME ADI-2 Pro has a more real-life sound at a downsampled 176 kHz while bit-perfect 352 kHz source material creates a bigger soundstage but with a slightly artificial sheen. This could be due to a weakness in the RME or something else in my system, or could be a system-matching issue. But listening at 176 kHz or 192 kHz is best.

    The AudioQuest DragonFly Cobalt can be amazing at 88 kHz or 96 kHz especially on MQA, provided you feed it quality power via a dedicated circuit ( such as the iFi iDefender+). Plugged directly into a Raspberry Pi USB port, the Cobalt’s 88 kHz or 96 kHz performance falls apart, to the extent that it sounds better at a downsampled 44 kHz or 48KkHz.

    The idea is to find the best sampling rate for each DAC in your system. Roon’s settings menu is easy to use and quick enough to do A/B comparisons. It is best to use a 352 kHz or 384 kHz track so you can try it at bit-perfect, 176 kHz or 192 kHz and 88 kHz or 96 kHz. Upsampling is rarely recommended, because it makes everything work harder to provide no additional musical information, but why not try it just to make sure? If you prefer DSD DACs, Roon can convert PCM to DSD on the fly and can even convert between DSD rates.

    Go ahead and spend an afternoon finding the sweet spot for each DAC. You may find your DAC never sounded so good. Once you have set up each endpoint/DAC, it will automatically receive its favorite format regardless of what track you play. Roon performs DSP in 64-bit floating point on the fly and if you keep to power-of-two sampling rate conversion (sampling rate conversion in multiples of two, I doubt whether you will worry too much about any quality reduction.


    LampizatOr Super Komputer music server.

    LampizatOr Super Komputer music server.


    Computer Networks and Switches Can Matter

    If you stream Tidal, it goes without saying that the best internet service available for your house should be installed. In my case it is 1000M fiber to the home. Use a quality router with plenty of processing power. Gaming or SOHO (small office/home office) routers start at $200, which is a lot for a home router but cheap for audio.

    Computer audiophiles try to put their audio devices on a separate LAN on the home network to avoid as much as possible dear son’s YouTube or World Of Tanks running on the same piece of wire as your music and using up data resources. The best implementation is to join the audio LAN to the rest of the network as near to the main router as possible.

    The audio LAN should use a network switch that is known to work well for audio, because the switch has an audible impact. The D-Link DGS-108 is widely accepted by audiophiles. Better audiophile-approved switches (often based on the DGS-108) are available but the more expensive models may be overkill for systems costing under $10,000. Gaming switches, although high performance, are not automatically better and can sound worse in my experience. In all cases, replace the included wall wart power supply with something better.

    As with USB cables, network cables matter and I and others find that the use of different cables can be clearly audible. However, more expensive is not automatically better. In the systems with which I have experience, cat6 from a quality manufacturer such as 3M beats cat7. This is an area that needs investigation. If you go with 3M cat6 round cables, I doubt you will go far wrong. Avoid flat cables as they generally have inferior sound.

    Do Not Rule Out Wi-Fi

    Engineers (including myself) will tell you that wired LAN is preferred over Wi-Fi for critical networking applications due to faster, more consistent and more reliable data transfer. But high-end audio does not always follow scientific theory.

    In my system, Wi-Fi surprisingly gives better sound quality. Both my Ethernet and Wi-Fi are the best available for home use. One explanation is that since there is no longer any physical connection between the (noisy) computing devices and the (quiet) audio devices using (wireless) Wi-Fi , perfect isolation has been achieved. The science does not really matter to me because I know what I hear. If your endpoint streamer has Wi-Fi, why not give it a try?


    If you have gotten this far – congratulations! You have a fully working high-end computer audio-based system. On hi-res material, it has the best sound you have heard in your room. When listening to CD-quality (16-bit, 44.1 kHz) material you are matching the sound quality from your CD transport. You have managed to keep most of your existing equipment and you carried out as much computer engineering as you care to enjoy, or avoid.

    You succeeded because you relied on a background of years of experience in high-end audio and on the only measuring instrument that really matters: your ears.

    Chat groups are full of keyboard warriors who do not want to invest the effort or do not have the means to play in the high-end arena. Professional engineers understand that science and theory are merely a way to approximate the real world. If the world always worked according to science, all products would be perfect and there would be no need for prototyping or testing, or improvement.

    No one doubts that Boeing engineers know aircraft science and technology like the back of their hand. Yet the fact that they still test their airplanes rigorously shows the acceptance by even the best engineers that science only gets you part of the way. Real-world experience will tell the rest.

    Welcome to the world of high-end computer audio.

    Header image: RME ADI-2 Pro AD/DA converter.

    10 comments on “Moving to Computer Audio: A Comprehensive Overview, Part Two”

    1. Super stuff Ed thank you.
      Some be of us have dived into computer audio find it the best sound quality we have ever had.
      And that the “Final five percent” though worthwhile is not a reason to just jump in.

      We have to get out that message. Start with the defaults .

      *Just do it*

      Then tweak if you must … because we can’t help ourselves. Chasing That final five percent.
      Guilty, Your Honour. But it’s not compulsory not at all.
      Enjoy the Music and Appreciate the Benefits of Metadata.

      In Zen and the Art of Motorcycle Maintenance he understood our people:
      Some believe they they can take on a new problem and probably overcome it.

    2. Good morning Edd!
      I am a blind Audiophile man.
      For right now, I’m running an HP Pro3500 Series desktop computer with Windows10Pro, build 2004.
      Because of the fact that I’m blind, I have to use screen reading text to speech software like JAWS.
      Someone at House Of Stereo in Jacksonville Florida, suggested to me, jRiver.
      But the problem I have with it, is, I couldn’t install it.
      I have been checking out quite a few dacs.
      The first one, was the Korg DSDac10-R.
      This one, can allow you to make and playback DSD audio files.
      But I’m not so sure of the software that you have to both download, and install.
      The second dac that I thought about because of the fact that it uses a cupple of tubes on the outputs, is the Black Ice Audio FS-DSD WIFI.
      It too can play DSD files from any computer via USB.
      I wanted a Ps Audio Sprout 100, but they’re out of those right now.
      But the suggestion that was made to me, was the Ps Audio Stellar Strataa.
      But again, I don’t know if that will work for me or not.
      I’m wanting to put DSD in, and get DSD out without any PCM convertions.
      Do you have any ideas?
      Which dac should I really go with.
      And also, is there something that’s easier to use other then jRiver and Roon?
      Thanks in advance!

      1. Hi John, JRiver is an excellent software. Are you in a position to describe the situation, perhaps a reader who is currently using JRiver can help? Also does anyone know whether JRiver can be set to output everything in DSD format? Thanks.

        1. Hi again Ed!
          Here it is!
          First, I done a Dogpile surch for jRiver.
          I instructed my computer to run it.
          It got me as far as the install wizard.
          It got me to the next Butten, I hit it.
          But the thing just stood still.
          This is the reason why I couldn’t install jRiver.
          JAWS quit reading it to me after that.

        2. Good afternoon Ed!
          Did either you or someone ever get the chance to figure out the jRiver instillation problem?

          Remember, that JAWS quit reading the install wizard out loud to me.
          This is the reason why I couldn’t install jRiver on my computer.

    3. Excellent article! I especially like the last 2 paragraphs about science and engineering. How true. I have a great deal of faith in science and engineering, but I’m also aware that we don’t know everything. The final judge of sound quality should be your ears, not numbers and specs. Not that specs aren’t useful, I use them to compare things all the time, but they aren’t the whole story.

    4. This is a great overview of computer audio, thanks.
      I’m a Roon nut for two years now. I love it as my main interface/controller to stream Qobuz and play local files, many which are DSD. I have a DirectStream DAC and have been using the Bridge II which has worked flawlessly for 3 years with REALLY good sound quality. Alas, I got the itch for “better” SQ and replaced the Bridge II with: (1) a Sonore ultraRendu (with their cheap LPS), and (2) a Sonore ultraDigital (with their cheap LPS) for converting to I2S then into my DAC. Everything is plugged into a PP12. It sounds so good, I’m selling my turntable. It sounds stupid-good, a lot better than the Bridge II. (I haven’t even begun to upgrade my router and switch, but that’s next.) I still play my favorite music on SACDs via a Memory Player, but honestly, the SACD/DSD layer barely sounds better than Qobuz. What DOES sound better are pure DSD files (e.g., from Octave Records or Blue Coast). The DSD file detail and dynamics are just astonishing.
      Check out Sonore. They are a good value and they have solutions tailored to PS Audio’s beloved I2S connection — including an end-game fiber optic-based streamer/isolation kit.
      Running Roon/Qobuz on an iPad is just remarkable. I have access to pretty much all the music ever recorded and wonderfully-tuned algorithms for learning new music. It’s a great time to be a computer audio nut.

    5. “Upsampling is rarely recommended, because it makes everything work harder to provide no additional musical information…”

      The above statement makes as much sense as saying audio somehow skirts around the physics and is it’s own parallel universe of critical arrributes. Digital audio makes errors even LESS a problem than analog as errors and noise are expunged through the digital process with bit perfect signals going around the world. In God we trust all else bring the data.

      Changing a system can change the process, but without the data to explain why there can be no trust in the system. The system is running YOU, verse you running IT. Audio in general needs to realize this and continue to build TRUST into the system of understanding instead of constantly leaning away from it.

      Back to upsampling. The sound of your system is the AD and DA filter errors in digital audio…and this is where the actual DATA shows the problems to exist. If someone can show me a BER, Bit Error Rate, test of the rest of the chain I’m all ears to see the errors as the DAC receives them.

      ALL digital is an “error” from a perfect square wave, but the system allows errors of a specific magnitude such that the original data can be clocked perfectly at the NIC end points. As long as the error is within Ethernet guidelines the perfection is exactly the same on BER.

      Using on the fly upsampling (modern CPU’s inhale this conversion process) can allow you to use the best sounding Digital to Analog section in your DAC this is of the highest benefit to you. It as critical as the speakers you use. There is a significant advantage to trying it and supported with the filter science.

      Changing the firmware in your PS Audio Direct Stream DAC and the test data shows you WHY this is so. J.River MC28 DSD upsampling leaves Native DSD untouched where PCM can be upsamples to DSD rates on the fly to the best your DAC accepts. Well, if you use a DSD DAC.

      There is true scientific data that explains why DSD512 or even 1024 yields a more linear filter through audio than lower rates. Correct, the data is the same, but the data rate changes the analog filter linearity. It also explains why it is filter, filter, filters in digital to analog music or even QOS, quality of service, in a phone call. Same exact stuff.

      The signal path ALWAYS follows the science. When we can better model that science the more trust we have in making true changes. When the changes are made, and we know we made them, are we getting a benefit? If you can’t show me a true change to use as a REPEATABLE reference to evaluate what’s the point? I want to pay for CHANGE, not an opinion.

      High end audio gets stuck in that use your ears excuse to side step any true data showing CHANGE, not that it is going to be better if you change something, we change it back after all. But we know the reference between the changes and can repeat them anywhere. WE control the system.

      There are, as Ed shows, different conveniences in moving digital data. To better control the system like it or not we need the true DATA to weigh what is better, if any are truly better, at proving a bit perfect solution. Modern Ethernet is so robust errors are near a thing of the past and I’d say are a thing of the past with a properly measuring channel inside the Ethernet guidelines.

      The comment on FLAT cables is not “true” as it isn’t supported with data. My comment is supported with the data. A flat cable, like Belden Mediatwist, REMOVES a great deal of pair-to-pair NEXT as the adjacent pairs (pairs next to each other) are the prime offenders in energy transfer between pairs. The alternate pair NEXT, or pairs with a pair between them or two more, use good old distance to mitigate the coupling and thus are quieter. We have just three “close” pairs in a flat design. The other three combinations are physically far apart.

      The Mediatwist design also allows LONGER lays sets to be used lowering attenuation and with lays optimized for better adjacent pair NEXT and thus a higher ACR, attenuation to cross talk ratio than a round cable. This improves the Shannon’s law bandwith. And no, it isn’t a feeling it is in the actual test DATA it is a better design where a flat cable can be used for CAT6. No round cable can side step the physics of the geometry of the EM fields.

      I’m all for digital audio, but is does and will ALWAYS follow the science. When we don’t understand the science that doesn’t make it go away. Ignorance just allows the science to control us in a non repeatable system of errors or accidental and random improvementd. But to be fair to ourselves, let’s get down to at least finding the errors repeatably and then remove the causes.

      I use J.River MC28 on an AMD 5600G CPU ITX system to on the fly USB stream DSD512 as it allows the best sounding filter to be used. I listened to all the filter analog errors, not digital as I’m bit perfect to Ethernet standards, and went the way that is most often colored to my liking once the source and filter errors are superimposed one on top of each other.

      Great article from Ed to show your options on how to move your music digitally but we need to accept that the science is REAL and nothing avoids that. That’s good as we can, in time, drill in and repeatably build better and better components for everyone to enjoy.

      Not all true changes can be heard by everyone and this is good…we get the sytem transparent to the process with measurements. No evidence of ANY repeatable change is hard to hear and even worse, get me to pay for.

      When we listen to a DAC, and change the DAC for the same DAC, the sound should follow. THAT is the science working. Change to a different DAC and again, the science follows the changes made. As long as we repeat the design, the science stays the same and does not ever suggest it somehow doesn’t matter. Like it or now we listen to the science ALWAYS.


    6. I think all would recognise that solid science is essential to making any decent engineered device. But I don’t think it’s anti-science to think there’s more to audio than we currently understand or can measure. A problem I have is that theoretical and measured non-linearity of modern hi-end electronics (both digital and analog) is already vanishingly low such that we are repeatedly told it is inaudible. And yet these differences are readily apparent to audiophiles. The result is that many of us are forced to rely on (non-scientific?) frameworks while the science catches up.
      Happy listening!
      Best regards

    Leave a Reply

    Also From This Issue

    Dogged Determination

    Lyle Mays – Composer, Arranger, Producer and Keyboardist, Part Two

    Part One (Issue 144) covered Lyle Mays’ recordings as a…

    Stuff Smith: Black Violin

    Leroy “Stuff” Smith, born in 1909, helped turn the violin…

    Dar Williams: Writing Songs That Matter

    The folk community is quite a bit like the jazz…
    Subscribe to Copper Magazine and never miss an issue.

    © 2023 PS Audio, Inc.

    linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram