Computer to speakers: getting the best sound, and the traps along the way.

For the best sound from a computer, get the digital audio out of the PC unchanged and let one good converter (DAC) turn it into sound: either your amplifier’s own DAC, fed by a USB-to-optical or coaxial card, or an external USB DAC. Then check four things that quietly undo it. Windows’ mixer can resample and reformat everything (set the device’s Default Format, or use a player’s exclusive mode). Effects and “enhancements” can be switched on without you knowing. A 96 kHz light only proves the sample rate arrived, not that the music is genuinely hi-res or that 24 bits got through. And loud modern songs can clip inside the DAC even when playback is bit-perfect. Do those and the computer stops being the weak link.

The chain, and the two good ways to build it

Every sound you hear from a computer passes through the same steps:

1. The file, say a FLAC rip of a CD (16-bit, 44.1 kHz) or a hi-res download (24-bit, 96 kHz).
2. The player, which decodes it into plain numbers called samples.
3. Windows, which either mixes those samples with every other program’s sound (shared mode) or hands the device to one player on its own (exclusive mode).
4. The output device: a converter (DAC) that makes sound, or a digital output that passes the numbers on to one.
5. The cable to the amplifier: digital (optical or coaxial) or analogue (RCA).
6. The amplifier and speakers.

The DAC, the digital-to-analogue converter, is the step that turns the numbers into a voltage that moves your speakers. Everything before it is just moving numbers around, and the goal is to move them without changing them. There are two good ways to do it:

Route A: let the amplifier do the converting

This is what we use. A DSD TECH SH-AU02A, a small USB card built on C-Media’s CM6542 chip, takes the audio over USB and sends it out as S/PDIF, the digital link on optical (TOSLINK) and coaxial sockets (more on the plugs). Our Marantz AV receiver takes it in over a coaxial cable and does the converting itself.

Done properly, nothing in the computer touches the sound at all: the same numbers that are in the file arrive at the receiver’s DAC. Coax and optical carry exactly the same signal. Optical has one extra trick, which we haven’t needed ourselves: it’s light, not electricity, so there’s no electrical connection between the PC and the amp, and computer noise and ground-loop hum can’t travel along the cable.

The catch is the card. A USB-to-S/PDIF card has one job, and cheap ones don’t always do it at full quality: some chips are 16-bit only, or stop at 48 kHz. Ours genuinely does 24-bit at 96 kHz, but only if the player asks for it in exactly the right way, which is the first trap in our story below.

Route B: an external DAC

We haven’t tried this route ourselves yet, so this part is general advice rather than bench notes. An external USB DAC does the converting itself and sends ordinary analogue sound to your amp’s line input. It’s the natural choice when your amp has no digital inputs, or when you want a better converter than the one built into the amp. Many modern USB DACs follow USB Audio Class 2, which Windows has supported with its own built-in driver since Windows 10 version 1703, Windows 11 included, and they often go well past 96 kHz.

Two things to know. Many USB DACs have their own hardware volume control, which Windows can drive; that’s handy, but it means the DAC’s volume and the amp’s volume both matter, and you want one of them doing the work, not both fighting. And if your amp is an AV receiver, its analogue input may not stay analogue: a receiver can convert it to digital to apply its own processing (bass management, room correction, surround modes), then convert it back. A Direct or Pure Direct mode switches most of that off, and on some receivers keeps the signal analogue all the way (Yamaha’s manual says so plainly; many don’t say). Our Marantz’s manual only says its Source Direct mode bypasses the tone controls and bass management. That’s not necessarily bad, but if the receiver’s converters are in the path anyway, that’s half the reason you bought the external DAC.

And the routes we’d skip for serious listening

• The PC’s own headphone or line-out jack. It’s a DAC too, just a cheap one sitting next to a power supply and a graphics card. It can be perfectly fine. Turn the amp up with nothing playing: if you hear hiss, buzz or whine that changes as the computer works, the noise is coming from inside the PC.
• Bluetooth. It’s convenient, but the sound is almost always squeezed through a lossy codec on the way (SBC, AAC or aptX). Sony’s LDAC carries more, up to 990 kbps, but that’s still less than uncompressed CD audio (1,411 kbps), let alone a hi-res file. Qualcomm’s aptX Lossless can carry CD-quality sound unchanged when the signal is good, but only if both the computer or phone and the headphones support it. Either way, your hi-res FLAC isn’t arriving as it left.
• HDMI to a TV or monitor. It works, but the TV usually does its own processing, and a monitor’s built-in speakers are an afterthought. (We do use a monitor’s speakers in one way: turned to zero, as a safe output for testing new playback features. More on why below.)

What our own setup taught us

We’re building Ironbark, our own music player for Windows. It isn’t publicly released yet. One of its jobs is bit-perfect playback: handing the sound device exactly the numbers in the file, and proving it by comparing what was sent with an independent reading of the file. Testing it on our own Route A setup turned up three lessons we didn’t expect.

Lesson 1: “24-bit” quietly became 16-bit

On 12/10/2026 John played a 24-bit, 96 kHz FLAC through the DSD TECH card in exclusive mode, and Ironbark warned that the device was taking 16-bit samples. It was right. The card accepts 24-bit audio only packed three bytes to a sample. Ironbark had asked for 24-bit in the more common 4-byte box, the card said no, and the sound library fell back to 16-bit without complaint. The music played normally, the receiver’s 96 kHz light came on, and the bottom 8 bits of every sample were gone.

The fix (Ironbark 0.46.1, the same day) was to check the format the device actually agreed to, not the one we asked for, and try packed 24-bit when the first answer falls short. We’ve written up the whole story, and every place along the chain where bits can be lost, in Why a 24-bit file can quietly play as 16-bit.

The lesson for everyone: the format the file has, and the format the player asked for, are not proof of what the device is getting. Look for a player that tells you what it actually got.

Lesson 2: a sound card with no volume knob

The DSD TECH card has no volume control of its own; Windows reports none. That’s normal for a digital output, since its job is to pass the numbers on untouched. But it has a consequence that’s easy to miss: in exclusive mode, Windows’ volume slider can only work on a device with its own hardware volume (Microsoft explains). On a card like ours, the player’s volume is the only volume in the computer, and bit-perfect means the player’s volume is at 100%.

So the amplifier’s knob is the volume control, full stop, and the amp has to be turned down before the first song. A player at 100% into an amp left at a party level is a sudden, full-scale blast, which is hard on ears, headphones and tweeters. We take it seriously enough that Ironbark asks before turning the volume to 100% on a device like this, has a per-output “Volume on my amplifier” setting (0.47.0) for exactly this card, and that we test new playback features on a monitor’s speakers turned right down, never on headphones.

Lesson 3: bit-perfect can still clip

This one surprised us most. A recording is stored as samples, and no sample can go above full scale (0 dBFS, the loudest number the file can hold). Many modern masters are made so loud that their samples sit right up against that ceiling. The trouble is that the DAC doesn’t play the samples as steps; it draws the smooth wave that runs through them, and between two samples at full scale, that wave can bulge above full scale. These are called inter-sample overs, or true peaks above 0 dBTP.

They’re common. In 2025 the audio blogger Archimago scanned more than 107,000 tracks in his family’s music library: 44% of the rock tracks had true peaks above +0.1 dBTP, and the worst track, an electronic one, peaked about 7.7 dB above full scale. To cover 99% of tracks, a DAC would need about 2.4 dB of spare room for rock, 2.9 dB for pop and 4.2 dB for electronica. A DAC built with no room above full scale chops those bulges off, which is clipping: a brief burst of distortion on the loudest moments. Some converters are designed with a few dB of spare room for exactly this (Benchmark builds 3.5 dB above full scale into its DAC2 and DAC3); many ordinary ones have none. When Archimago measured three DACs in 2018 (all models from around then, so newer ones may differ), the TEAC UD-501 had no spare room, the RME ADI-2 Pro FS about 2 dB, and the Oppo UDP-205, a disc player with its own DAC, an estimated 3.5 dB.

And here’s the twist: bit-perfect playback delivers the clipping perfectly. The numbers arrive untouched, and a DAC with no spare room clips them anyway. Unless your DAC has that room, the cure is to turn the song down a little before it, which, by definition, is no longer bit-perfect.

That’s how Ironbark’s clipping guard works. It measures each song’s true peak ahead of time, by rebuilding the wave between the samples at 8× the sample rate. That’s the method in the international true-peak standard (ITU-R BS.1770), which asks for at least 4× and says more is better. Then it turns down only the songs that need it, by just enough. Every other song stays bit-perfect. If your DAC is known to have room above full scale, a “My DAC has headroom” setting tells the guard how much. When bit-perfect and the guard disagree, the guard wins, and the player says so plainly: the song “isn’t bit-perfect (the guard comes first).”

We thought hard about that order, and we’d make the same call again. Turning a 24-bit stream down by 2 or 3 dB costs nothing you could ever hear: the change sits far below the music. A DAC clipping the loudest moments is a real, if small, distortion. Clean beats bit-perfect. The one exception is music with something hidden in the bits, such as a DTS CD or an HDCD disc: any change in level breaks those, so they need the path left completely untouched.

In fairness, here’s the honest part. When John first sat down with our clipping demo files, loudness-matched so neither version won by being louder, he couldn’t hear the difference. So instead of asking people to trust their ears, Ironbark shows it: its Clipping window draws the wave live, red wherever it passes full scale, and lists every over in the song with a click to jump there. (There’s an Experimental “Hear it” switch to compare an ordinary DAC with a clean one, A against B, if you want to try your own ears.) Clipping on a good system is usually subtle. It’s still worth removing when it costs nothing to remove.

The traps, and how to check each one

1. Windows’ mixer (shared mode)

Normally every program shares the sound device and Windows mixes them together. In shared mode the device runs at its Default Format (Microsoft’s explanation), so anything at a different sample rate is resampled on the way, by Windows or by the player. Many devices come set to 48 kHz, which means every CD rip (44.1 kHz) gets resampled on its way out. Done well, resampling is very hard to hear, but Windows’ own resampler isn’t the best: measurements on Windows 10 in 2015 found it trades quality for speed. Either way it’s a change, and one you didn’t ask for.

Check: press Win+R, type mmsys.cpl, select your device, Properties › Advanced. Default Format is what Windows sends in shared mode. When we set up the DSD TECH card, we had to raise ours from the setting it came with, so it’s worth looking even on new hardware.

Fix: set it to 24-bit at the sample rate of most of your music (44.1 kHz for a CD-rip library), or use a player’s exclusive mode, which skips the mixer and switches the device to each song’s own rate.

2. The 96 kHz light that proves less than it seems

Your receiver’s 96 kHz light means a 96 kHz signal arrived. It doesn’t mean the music is hi-res. Three ways it can mislead:

• Windows is upsampling. Set the Default Format to 96 kHz and everything arrives at 96 kHz, CD rips, YouTube and system sounds included. The light glows for all of it, but what’s arriving is Windows’ resampled version of a 44.1 kHz file, not hi-res, and not what’s in the file. Upsampling can’t add detail that wasn’t recorded. (Some listeners like a player’s upsampling for its filter, which is a different argument: it changes how the wave is rebuilt, not what’s in it.)
• The file itself is upsampled. Some “hi-res” downloads were made by converting a CD-quality master up to 96 kHz. One clue is in the spectrum: genuine hi-res usually has some content above 22 kHz, while a file made from a CD stops sharply around 20–22 kHz (or about 24 kHz if it was made from a 48 kHz master). Ironbark’s file Check flags this as a clue (never as “fake”), as do other spectrum tools.
• Room correction can quietly undo it. In receivers that use Audyssey room correction, the correction runs at no more than 48 kHz (the makers’ choice, to save processing), so with it switched on a 96 kHz signal is brought down to 48 kHz inside, whatever the display says. A Direct or Pure Direct mode, which switches room correction off, avoids it.

And as Lesson 1 showed, the light says nothing about bit depth. Many receivers simply don’t display it.

3. Effects you didn’t switch on

Windows and sound-card drivers can add processing: audio enhancements, spatial sound (Windows Sonic and others), loudness equalisation, and makers’ own bundles from Realtek, Dolby, DTS or Nahimic. Some are on by default on new laptops and PCs. Any of them changes the sound before it leaves the computer.

Check: in the same Properties window, look for an Enhancements tab and switch them off. In Windows 11, go to Settings › System › Sound, select your device, and set Audio enhancements to Off; on the same page, check Spatial sound is Off. On the Communications tab of the Sound control panel, set “When Windows detects communications activity” to Do nothing, or your music will dip whenever a call app thinks you’re on a call (out of the box, Windows turns other sounds down by 80%).

The surest check is to measure. In shared mode, Ironbark’s Check what Windows does to the sound records what Windows actually sent the device and compares it with what the player handed over, and says whether Windows left it untouched, resampled it, changed its level or altered it with effects. Exclusive mode skips Windows’ own effects by design, though effects built into the sound hardware itself can still apply.

4. 24-bit becoming 16

Covered in Lesson 1. Check the Default Format offers a 24-bit option for your device; in exclusive mode, use a player that reports the format it actually opened the device with. A DTS music CD or HDCD disc is a good test that a 16-bit path is untouched (start with the amp turned down: if anything alters the bits, a DTS disc plays as loud white noise), but it can’t show whether 24 bits get through. The details are here.

5. Volume in the wrong place

There are up to four volume controls in the chain: the player, Windows, the DAC (if it has one) and the amplifier.

• In shared mode, any volume below 100% in the player or Windows is a digital change to the samples. At 24-bit that’s harmless for listening (there’s plenty of room below the music), but it isn’t bit-perfect.
• In exclusive mode, set the player to 100% and use the amplifier. On a device with no volume of its own, like our card, turn the amp down first (Lesson 2).
• Don’t stack them. Volume set very low in the computer and very high on the amp raises the noise floor; the reverse, on Route B, can overload the amp’s analogue input, since some DACs put out more than an amp expects. Aim for the computer at or near full, and the amp doing the work.

6. Loud songs clipping in the DAC

Covered in Lesson 3. Check: if very loud tracks sound harsh at their peaks and quieter recordings don’t, inter-sample overs are one possible cause. Often, though, the harshness is clipping already baked into the master, and no setting can undo that. A player with a true-peak meter or clipping guard will tell you which; failing that, turning the player down by 3 to 4 dB on those tracks is an easy test, and at 24-bit it costs you nothing.

7. Switching sample rates

In exclusive mode, the device switches to each song’s own sample rate, which is what you want. But when an album at 44.1 kHz is followed by one at 96 kHz, the receiver has to re-lock to the new rate, and many mute while they do: Denon and Marantz warn in their manuals that the sound can cut for 1 to 2 seconds when the sample rate switches. You can lose the start of the next song, and two songs at different rates can’t play gaplessly into each other. We haven’t checked whether our Marantz does it. It’s a harmless quirk once you know it’s normal; it isn’t a fault in the card or the player.

8. The cable and the connection

Digital cables are forgiving: either the numbers arrive or you hear obvious faults, clicks or drop-outs. Audiophiles argue about jitter (tiny timing wobbles in the signal), but careful measurements have found no difference between cables, cheap or dear. Optical and coaxial carry the same S/PDIF signal. A few practical points:

• Optical can’t carry ground-loop hum along the cable. Don’t bend the cable sharply, and push the plug fully home (it clicks).
• Coaxial uses a 75-ohm cable; an ordinary RCA lead usually works over short runs.
• S/PDIF carries stereo PCM. Surround over optical or coax has to travel as a compressed Dolby Digital or DTS stream, which doesn’t matter for stereo music.
• Analogue cables pick up hum more easily. If you hear hum on Route B that stops when you unplug the USB cable from the PC, it’s probably a ground loop: a USB isolator, or a ground-loop isolator on the analogue lead, usually cures it. Never remove a mains earth to stop hum.

A ten-minute check for your own setup

1. Turn the amplifier down. Always the first step.
2. Windows › Sound control panel (Win+R, mmsys.cpl) › your device › Properties:
   • Advanced: Default Format at 24-bit and the sample rate of most of your music; tick Allow applications to take exclusive control.
   • Enhancements / Spatial sound: off.
3. Communications tab: Do nothing.
4. Player: turn on exclusive or bit-perfect mode if it has one, volume at 100%, and look at what it says it opened the device with: rate and bit depth. Be wary of one that only repeats the file’s format back to you.
5. Play a CD rip, then a hi-res file. Watch the receiver’s sample-rate display change (44.1, then 96). If it says 48 or 96 for both, Windows is resampling: you’re in shared mode.
6. Bring the amp up slowly to your normal level.
7. Play your loudest album. If the peaks sound harsh, try the player 3 to 4 dB down on those tracks, or switch on a clipping guard if your player has one.
8. Listen for noise with nothing playing, amp at your usual level. Hiss or buzz that changes with the computer’s workload means noise is getting in: try the optical route or another USB port.

Does any of this matter?

Honestly, less than speakers, the room and the recording do. Moving your speakers, or choosing a better-mastered release of the same album, will change what you hear far more than anything on this page.

But the computer shouldn’t be making things worse, and every trap above is something we found, or nearly missed, on our own bench, with good equipment and lossless files. The fixes are free: a few settings, the right mode in the player, and the amp turned down before you press play. Once the numbers arrive unchanged, and clean, at a good DAC, the digital side is done, and you can get back to the bit that really matters: listening.

The card we used

USB sound card (optical and coaxial out)

DSD TECH SH-AU02A

What we use it for: sending music from a Windows PC to our Marantz AV receiver over a coaxial digital cable (Route A), and testing Ironbark’s bit-perfect mode on it.

Chip

C-Media CM6542 (named in DSD TECH’s own listing): USB Audio Class 1, S/PDIF out at up to 24-bit, 96 kHz

Connects by

USB, which also powers it

Outputs

Optical (TOSLINK) and coaxial S/PDIF, plus a 3.5 mm analogue output (per the listing)

Listed as

Up to 24-bit, 96 kHz

What we measured

In exclusive mode, at 96 and 44.1 kHz: 24-bit only as packed 3-byte samples. Asked for 24-bit in a 32-bit container, it fell back to 16-bit.

Volume control

None of its own: Windows reports no hardware volume, so set the level on your amplifier

Driver

Windows’ built-in USB Audio driver; nothing to install on our Windows 11 PC

Know before you buy: it does 24-bit, but a player has to ask for it the right way. In shared mode, check the Default Format in Windows. In exclusive mode, use a player that tries packed 24-bit and reports what it got. Because it has no volume control of its own, turn your amplifier right down before the first play in exclusive mode.

See it on Amazon AU (paid link)

Common questions

Sources

Microsoft

• Device Formats (shared mode, the Default Format, exclusive mode), Microsoft Learn. link. Checked 12/10/2026.
• Exclusive-Mode Streams, Microsoft Learn. link. Checked 12/10/2026.
• Endpoint Volume Controls (hardware vs software volume in exclusive mode), Microsoft Learn. link. Checked 12/10/2026.
• USB Audio 2.0 Drivers (Windows’ built-in driver, since Windows 10 version 1703), Microsoft Learn. link. Checked 13/10/2026.
• Low Latency Audio (exclusive mode bypasses the audio engine), Microsoft Learn. link. Checked 13/10/2026.
• Default Ducking Experience (the Communications tab, 80% by default), Microsoft Learn. link. Checked 13/10/2026.
• Fix sound or audio problems in Windows (Audio enhancements, More sound settings, Default Format), Microsoft Support. link. Checked 13/10/2026.
• How to turn on spatial sound in Windows, Microsoft Support. link. Checked 13/10/2026.

Standards

• Recommendation ITU-R BS.1770-5, Algorithms to measure audio programme loudness and true-peak audio level, ITU. link. Checked 13/10/2026.
• Digital audio interface – Part 3: Consumer applications (IEC 60958-3), IEC. link. Checked 12/10/2026.
• Digital audio – Interface for non-linear PCM encoded audio bitstreams applying IEC 60958 – Part 1: General (IEC 61937-1), IEC. link. Checked 13/10/2026.

Manufacturers and measurements

• DSD TECH SH-AU02A listing, with DSD TECH’s model comparison (main chip per model), Amazon AU. link (paid link). Checked 12/10/2026; title re-checked 13/10/2026.
• CM6542 datasheet v1.93, C-Media Electronics. link. Checked 12/10/2026.
• SR4300 User Guide (Source Direct mode; 96 kHz PCM in Auto, Source Direct and Stereo modes), Marantz. link. Checked 13/10/2026.
• Enjoying higher-quality playback (Pure Direct), Yamaha RX300A online manual. link. Checked 13/10/2026.
• Connecting and playing back from a digital device (Coaxial/Optical), Denon PMA-1600NE manual. link. Checked 13/10/2026.
• Connecting and playing back from a digital device (Coaxial/Optical), Marantz PM7005 manual. link. Checked 13/10/2026.
• Intersample Overs in CD Recordings (John Siau), Benchmark Media Systems. link. Checked 13/10/2026.
• A look at the music library: How much DAC intersample overhead is enough? (true-peak scan of 107,000 tracks), Archimago’s Musings. link. Checked 13/10/2026.
• A look at Intersample Peaks and Overload Tolerance, Archimago’s Musings. link. Checked 13/10/2026.
• GUEST POST: Why We Should Use Software Volume Control / Management (Bennet Ng), Archimago’s Musings. link. Checked 14/10/2026.
• MEASUREMENTS: Windows 10 “Audio Stack” / DirectSound Upsampling, Archimago’s Musings. link. Checked 14/10/2026.
• Denon AVR-X3700H AVR Review (Pure Direct and analogue inputs; receiver DAC measurements), Audio Science Review. link. Checked 14/10/2026.
• Audyssey MultEQ Room Correction Interview With Chris Kyriakakis (Audyssey run at up to 48 kHz in AV receivers), Audioholics, 2014. link. Checked 14/10/2026.
• MEASUREMENTS: TosLink digital optical audio cables (jitter, cheap vs glass), Archimago’s Musings, 2013. link. Checked 14/10/2026.
• LDAC developer site, Sony. link. Checked 13/10/2026.
• QCC30xx Series Bluetooth Audio SoCs for True Wireless Earbuds (aptX Lossless, 16-bit 44.1 kHz), Qualcomm. link. Checked 13/10/2026.

Our own

• Why a 24-bit file can quietly play as 16-bit (the DSD TECH story in full), Digital Greenhouse. link.
• Ironbark notes and live checks, 04/10/2026 to 12/10/2026 (clipping guard, packed 24-bit, Volume on my amplifier).