Can cable sound be measured? A €2.800 USB cable put to the studio test.
Is there such a thing as cable sound, or is it just a narrative? We connected two USB cables, each costing €2.800, to our studio chain, measured the entire transmission, and then listened to 20 rounds of blind listening. Here are all the figures, all the limits – and what cannot be deduced from them.
Transparency notice (advertising): VBS Audio provided us with two cables of the model “Eigen” free as a product sample The materials were provided. No money changed hands, there was no compensation, and no agreement was made regarding the outcome. The manufacturer granted permission for the publication of photos, company name, and website and agreed to the procedure in advance. We consistently distinguish between manufacturer specifications, our own measurements, and subjective listening impressions in the text.
Cable sound Digital audio cables have been a contentious issue in the hi-fi and studio world for decades – and one that's almost always debated without numbers. The discussion is particularly heated when it comes to digital audio cables, such as USB, coaxial, Toslink, or AES. A manufacturer contacted us: their cables made an audible difference, especially at the digital level. We agreed – on one condition: we would measure the entire chain, conduct blind listening tests, and publish everything, regardless of the outcome. This report is the result, with all the figures, all the limitations, and everything else we learned. not can say.
Contents of this cable sound test
- The result in five points
- How it happened
- The product and the manufacturer's claims
- The four digital transmission paths in the studio
- The test bench: devices, signal path, fuses
- Digital transmission: what was already known
- The analog measurements and why there are five passes
- Silence, sine tone and sweep compared
- The counter-test to the digital original
- The blindfolded round: 11 to 9
- What that means – and what it doesn't –
- Methods and Data Appendix
- Frequently asked questions
Is there such a thing as cable sound? We measured it in the studio.
Cable sound refers to the assumption that different cables audibly alter the same signal path. In this series of measurements, this was not demonstrable. Under the documented conditions, neither a consistent subjective preference nor a uniformly better fidelity to the original sound emerged for either of the two cable configurations. Small numerical deviations were measurable; however, in this setup, they cannot be definitively attributed to the USB cables. Therefore, this finding pertains to two specific cables in a specific chain, but it does not constitute a general conclusion.
Cable sound in five points: the result
- Digital transmission: Initially, the data arrived flawlessly at the 24-bit level in the tested sections – with the test cable just as with the original cable. The USB error counters of the RME interfaces remained at zero during the documented endurance tests.
- Frequency Response: Furthermore, both states overlap across the entire audible range except for less than 0,0002 dB.
- Music signal: Furthermore, the residual waveform difference between the conditions is approximately -109,9 to -109,4 dBFS – practically on the same order of magnitude as the variation between two recordings the same condition.
- The only noticeable deviation: Only in the noise band during silence was the reference condition around +0,104 dB (left) / +0,116 dB (right) higher. In addition, there is only one silent scan per condition – this does not provide any information about the cause.
- Blind preference screening: Finally, 20 rounds, one listener, 11 winning votes for the VBS cable, 9 for the RME cableThe two-sided exact binomial test against 50:50 yields p = 0,824.
Why we go to this trouble: We work in mastering Every day, we encounter differences that lie at the edge of perception. Anyone who wants to separate claims from findings must measure and listen in blinded conditions – otherwise, the price tag will be a deciding factor. And most importantly, because it's almost always misquoted: A non-significant result is not proof of sound equality. The test examined a preference – not whether someone can reliably distinguish between the two states. These are two different questions, and we only answered one of them.
How it came about: an offer for a cable sound test
First, some background: The manufacturer contacted me – a small-scale manufacturer, not a mass-produced product. We spoke on the phone, he told me a lot about his cables and his company's history, and, as I recorded on tape immediately after the test day, "Also always touted as making a huge difference, especially in the digital realm, by using such cables.".
Why do we even test audiophile USB cables?
My answer was therefore: gladly, but properly. I offered a comprehensive test scenario, including a blog post and complete results, including technical measurements and blind listening sessions. Open-ended.
My voice note also explains why I participated at all, even though I am skeptical of such statements:
"Since I myself am still very skeptical of such statements, because so much is always said about it, these typical audio myths. And yes, now we really have a manufacturer who actually knows what he's talking about, and where we simply want to check it out and see: what really comes out of it."
Why we agreed to the cable sound test
That's precisely the point. The issue of cable sound has been debated for twenty years, almost always without any concrete data. However, we have a studio, measuring equipment, and the capability to conduct a clean, blind test. So we're doing it – just as we usually try to... To test common assumptions about studio work in practice.
The product: an audiophile USB cable from the manufacturer
First, the facts about the test subject. Everything in this section is... manufacturer's instructionsThey are not the result of our measurements, and we have not verified them – that is not possible with the methods used here.
| Model | VBS Audio Own |
| Length | 1,5 m |
| Specification | USB 2.0 high speed (480 Mbit/s) |
| List price | €2.000 base for the first meter; €2.800 for 1,5 m |
| Pricing logic | non-linear – each additional step of 0,5 m adds 40% to the initial length (2 m would therefore correspond to €3.920) |
| Sales | Exclusively direct, limited quantity, no distribution in Germany |
Price as of August 2026, written information from the manufacturer. Anyone considering purchasing should inquire about the current price directly with the manufacturer. vbsaudio.com.
What the manufacturer says about the construction
Conductor material, microstructure, geometry, shielding and connector termination are considered an integrated system Developed. Details regarding shielding, grounding, and internal construction are confidential and will not be disclosed; therefore, the evaluation should refer to the finished cable. The manufacturer also requests that the product be presented as a "precision calibration/reference system tool" and emphasizes "signal integrity in the time domain."
Fair assessment of these theses
- Our measurements neither confirm nor refute the mechanisms mentioned. Because we didn't perform any material analysis, any high-frequency measurements on the line, and no calibrated jitter measurements. Anyone wanting to test this needs a measurement setup that can do exactly that.
- Regarding time behavior: The devices in this chain operate asynchronously – the converter clocking comes from the device itself, not from the USB connection. The ADI-2 Pro ran on... Clock Source InternalThat's an argument, but not a measurement from us, and it doesn't rule out mass or shielding effects.
- To play: According to the manufacturer, the cables are already broken in, but the sound can "open up further." The cables had been installed in the studio for about two to three weeks on the day of the measurement – this is my own statement regarding usage, not a controlled verification. A proper before-and-after test would have to include an unaltered reference run; otherwise, you'd be measuring device temperature and contact condition. We didn't do that here.
Cable sound with digital audio cables: four approaches, four discussions
First, a clarification: People who talk about digital audio cables rarely mean the same thing. In the studio, four transmission paths operate simultaneously, and they differ significantly from each other technically. This is important because the results of this test not automatically transferred to each of these paths.
| send email now | Typical application | Clocking | What is being discussed here |
|---|---|---|---|
| USB | Computer to audio interface | asynchronous – the device clocks itself | Grounding, shielding, interference in the analog stages |
| AES3 / AES-EBU | Interface to converter or active loudspeaker | Clock signal embedded in the signal, 110 ohms balanced | Impedance matching, reflections due to incorrect cable type |
| S/PDIF coaxial | Consumer devices, older converters | Clock signal is embedded in the signal, 75 ohms unbalanced | Impedance, ground connection between two devices |
| Optical Toslink | Consumer devices, galvanic isolation | The clock signal is embedded in the signal. | Light scattering and slope steepness over long distances |
Cable sound in hi-fi and in the studio: the same debate, different conditions
The discussion about cable sound is primarily conducted in the high-end hi-fi sector, and there it usually revolves around speaker cables and power cords. Both are analog connections: Cross-section, length, capacitance, and inductance determine measurable electrical properties, and in the case of speaker cables, the behavior also depends on the impedance of the connected speaker. Whether this translates into a sonically relevant difference in a specific case is a separate question – but the physical basis for it exists.
With a digital connection, however, things are different. Here, no music signal is transmitted, but rather a data stream. A single cable in a streaming or studio system cannot therefore sound "warmer"—it can either transmit the data correctly, or it can introduce errors. That's precisely why we tested both levels separately in this review: first the data transmission, then the analog output.
Furthermore, the quality of the connectors must be considered: Poorly fitting or oxidized connectors create intermittent connections – and these manifest as dropouts, not as tonal characteristics. The connectors of the test samples fit snugly and are screwed in, which is a more practically relevant advantage in everyday studio use than any claim regarding sound quality.
Why cable sound is evaluated differently with USB
Consequently, the crucial difference lies in the rhythm: At USB in asynchronous mode The clock in the device determines the conversion time, not the incoming signal. With AES3, coaxial, and Toslink, however, the clock is embedded in the data stream itself – in these cases, the discussion about cable influences is technically better justified because impedance errors and reflections can blur the signal edges.
Therefore, our test focuses exclusively on the USB connection. It says nothing about cable sound quality with coaxial or optical digital audio cables. Anyone suspecting problems there should first check the impedance of the cable used – a 75-ohm video cable behaves differently on an S/PDIF input than any random RCA cable from a drawer.
And another point that is often missing in cable discussions: A defective or borderline cable can This can indeed have audible consequences – however, with digital transmission, this doesn't manifest as a "warmer sound," but rather as dropouts, crackling, or connection interruptions. That's precisely why we logged the error counters.
Instead of speculating about cables: Let us hear where your mix really stands.
We'll tell you exactly what reserves are in your production.
The test bench: where cable sound could originate
First, one must distinguish between two things: what was recorded and what was overheardThese are different approaches for us.
Recorded signal path of the analog measurements
Identical in both cable states. Only the two red-marked USB connections were changed – and both were changed simultaneously.
Computer and test object
Recording loop in the project
Eavesdropping path – separate from recording path
Clock chain: The ADI-2 Pro (internal) passes the clock signal to the UFX+, which runs synchronously at 44,1 kHz. The recorded path includes XLR connections, the patchbay path, and both converter stages of the ADI-2 Pro. This is explicitly not an isolated cable, converter, or jitter measurement.
The admission process
In the project, the music track first runs in Cubase to a group called "Analog". There, a External Effect (Analog chain) with send and return set to exactly 0,00 dB. The signal passes through the analog output of the RME ADI-2 Pro FS R Black Edition Beyond that, via XLR and a direct path from the Flock Audio patchbay (23 L / 24 R, without any inserted processors) back to the analog input of the same device and from there to the recording track "Final Mix" – 44,1 kHz, stereo, 32-bit float. The fact that the send and return are set to exactly unity gain is not a detail, but a prerequisite for any comparison – nothing other than consistent... clean level management in the signal chain.
The eavesdropping route
This one, however, runs completely separately: Cubase → Acourate Convolver (FIR equalization) → RME Fireface UFX+ → AES3 → left Dutch & Dutch 8C → AES-THRU → right 8C. The digital-to-analog conversion for monitoring therefore takes place in the speakers, not in the interface.
What exactly was changed – and why that limits the cable sound assessment.
Finally, both USB connections, namely together in one operation.
| Gerät | Reference state | Test condition | Please note: |
|---|---|---|---|
| ADI-2 Pro FS R BE | RME standard cable, USB 2 | VBS Eigen E1, USB 2 | clean cable comparison in the same mode |
| Fireface UFX+ | included RME cable, USB 3 | VBS Eigen E2 USB 2 | Cables and Switch operating modes here |
This is a conscious decision and at the same time the most important limitation of the structure: The UFX+ branch is a Practical comparison: Delivered state versus test cableThis isn't just a cable comparison. RME ships the UFX+ with a USB 3 cable – nobody would swap that for a USB 2 cable in everyday use just for the sake of comparison. The manufacturer would have preferred the same setup; we opted for the real-world configuration and are disclosing that here. The device's technical specifications are listed in the ADI-2 Pro FS R manual.
Consequence for the evaluation: Since both cables were replaced together, the analog recordings are compared. two combined system statesThis does not causally isolate a single cable or interface. The recorded path also includes XLR connections, the patchbay path, and both converter stages of the ADI-2 Pro. It is not an isolated converter, cable, or jitter measurement—similar to comparing... analog hardware versus plugins, where you always measure the entire chain and never just one component.
Beforehand, everything was backed up: project copy, Cubase configuration, patchbay and mixer files, screenshots of the initial state, and a manifest with SHA-256 checksums. The raw recordings were checked again after recording and were neither normalized, edited, nor overwritten.
Digital transmission: the precursor to every cable sound question
In fact, the first question had already been answered before the day of measurement: Will the data arrive intact? These tests were conducted before the measurement day and were already documented.
Bit test on the ADI-2 Pro
First, regarding the bit test: The ADI-2 Pro can check in its device menu whether a test signal is received with bit-perfect accuracy. The result was identical with both the reference cable and the test cable: 16-bit file "16 bits passed", 24-bit file "24 bits passed". With the 32-bit file, the device displayed... both In some cases, only "24 bits passed" was displayed. This documents the resolution limit of the tested playback path. no 32-bit pass and no cable faultWe have not separately demonstrated exactly where this limitation arises.
Digital loopback on the UFX+
In addition, a 24-bit test signal was recorded via an internal loopback in the interface mixer and compared sample-accurately:
| Cables | compared samples | Deviations at the 24-bit level |
|---|---|---|
| RME cable (USB 3) | 353.280 | 0 |
| VBS Own (USB 2) | 353.280 | 0 |
However, this is important for classification: it is about equality. at the 24-bit level in the tested sectionIn a pure floating-point comparison, a tiny residual signal of a maximum of 1,0 · 10⁻¹⁰ remains – far below one bit at 24-bit resolution; it can be attributed to the 32-bit float recording path. We explicitly do not claim to have byte-identical files.
Technical terms from the cable sound debate briefly explained
dBFS – Decibels relative to full digital output. 0 dBFS is the maximum; all values below this are negative. −110 dBFS is therefore extremely quiet.
Loopback – A signal is fed directly from the output to the input in the device to check the transmission.
Total harmonic distortion (THD) – the percentage of additional overtones that a device adds to the signal.
Sweep – a tone that slowly sweeps from low to high. This is used to measure the frequency response.
Jitter – tiny fluctuations in the clocking rate during the conversion from digital to analog.
Error counter and endurance runs
Furthermore, in two documented runs of approximately one hour each during playback, the diagnostic values of the interfaces were at zero at the end: no CRC5/16 errors, no CRC32 errors, the AES connection was consistently in sync at 44,1 kHz – in the reference state as with the test cable.
However, limits are part of the equation here:
- ZWEI than The runs remained incomplete (process problems; the program terminated at the end). Therefore, there were... not four completed endurance runs, but two.
- The monitoring script checks processes and selected system events every 30 seconds. no dropout detectorIt cannot prove anything between two samples.
- The event filter uses English terminology and was running on a German-language system. The reported "0 events" are therefore incorrect. no reliable negative finding.
- The files on which the loopback reports are based were in their documented storage location on the day of evaluation. no longer locatableThe reports exist, but the analysis of the raw data was not reproduced on that day. This is a gap in the archive – and we are acknowledging it instead of concealing it.
My own commentary on the tape belongs here, in the correct sharpness: "The bit tests have already shown complete bit equality. So everything definitely went well there, but there was no noticeable difference or anything like that. But we didn't expect any." More precisely: equality at the 24-bit level in the tested sections. Nothing more, but also nothing less.
The analog measurements: Cable sound in the recorded signal
The real question, however, is not whether USB packets arrive. The question is whether something different comes out of the converter at the end of the chain. So that's exactly what we recorded.
Specifically, the following were recorded for each cable condition:
- Music – a 60,666 second long excerpt on a continuous loop, five complete cycles
- Stille – approximately 30 seconds without a signal
- 1 kHz sine wave – 5 seconds
- Sweep – 30 seconds logarithmic from 20 Hz to 20 kHz, with 2 seconds of silence before and after, peak level −18 dBFS
Everything was identical in both states: same source, same levels, same chain. Then both cables were changed and the entire series was repeated.
Why five runs instead of one?
Otherwise, every number would be worthless. An analog recording setup never produces exactly the same file twice – converter noise, temperature, minute level fluctuations. If I want to know whether a difference between two cable states means anything, I first need a benchmark for it. how much the same state fluctuates with itself.
Therefore, each condition was initially recorded five times and evaluated in two directions:
- Repetition: Pass against pass within the same condition – that is the noise floor of the process.
- Cross-comparison: Each run of one condition against each of the others – 25 combinations.
The identical inner section from second 5 to 55 was analyzed in each case, i.e., without lead-in or loop edges. No level adjustment, no sound processing, no resampling, no shifting below a sample – the signals were subtracted from each other in their raw state.
Result with the music signal: no measurable cable sound
| Comparison | Residual signal (RMS of waveform difference) |
|---|---|
| VBS vs. RME, all 25 pairings, both channels | -109,88 to -109,38 dBFS |
| VBS against VBS (repeat of the same condition) | -109,92 to -109,57 dBFS |
| RME versus RME (repeating the same condition) | -109,85 to -109,55 dBFS |
| greatest level difference of the useful signals | below 0,000040 dB |
The crucial comparison: changing cables versus simply repeating the same task.
Value ranges of the waveform residual difference across all evaluated pairings, identical inner section from second 5 to 55.
Without this benchmark – the repetition of the same state – the number −109,6 dBFS would not be classifiable at all.
This is, incidentally, the central table of the entire test. Because the difference between the cable conditions lies in... of the same order of magnitude as the difference between two images of the same stateThe areas largely overlap, but they are not identical – the cross-comparisons extend somewhat further upwards.
On the other hand, that means notThat the recordings are identical. They are not, and that's not to be expected with analog recordings. This also doesn't mean that a residual difference at −110 dBFS is inaudible. We haven't calibrated an audibility threshold, and a difference level is not a measure of audibility.
Temporal stability – the jitter argument in cable sound
Because the word "jitter" is often mentioned in connection with digital connections, we examined the timing of the signals. Sixty individual tests (five take pairs × six windows × two channels) yielded results with a search range of ±4 samples. Everywhere an integer residual offset of zeroThe subsequent fine-tuning remained below 0,000024 Samples, the derived equivalent drift slope under 0,0000043 ppm.
Nevertheless, the following applies: These figures are model-based diagnostics – not a calibrated jitter measurement. They demonstrate that the comparison method does not require time correction. Anyone wanting a true statement about jitter needs a measurement procedure specifically designed for that purpose; we did not use one and therefore make no claims about jitter.
Phase and group delay: the time domain in cable sound
Addendum dated September 21, 2026: The manufacturer rightly pointed out after the release that our cable sound measurement in the sweep only captured the amplitude – phase and group delay were not evaluated. This was a real gap. It has been filled using the existing sweep recordings, without a new measurement run and without any changes to the signal chain.
The transfer function of one condition against the other, after subtracting the constant delay, yields: integer offset null constant residual delay in both channels 0,091 ns left and 0,090 ns right, residual phase over 20 Hz to 20 kHz within ±0,12 degreesThe group delay difference above 160 Hz is between 0,01 and 1,1 ns. The median magnitude of the transfer function is −0,0001 dB.
For context: A sample at 44,1 kHz lasts 22.676 ns. 0,091 ns is roughly one two-hundred-and-fifteenth of that – and about six times less than the known take-to-take time variation of the same chain, which can be up to 0,54 ns. Therefore, the same pattern applies in the time domain as everywhere else in this cable sound test: The difference between the cables is smaller than the system's variation with itself.
Grenzen dieser Nachmessung
It lies only one sweep per condition Therefore, there is no A/A reference for this measurement – the 0,09 ns cable difference and the 0,09 ns take variation are inseparable here. And the 20 to 40 Hz band is unusable: it yields 50 ns on the left and 5,9 ns on the right, contradictory values, because the logarithmic sweep has the lowest energy per frequency line in this range. This band is explicitly not shown as a result.
QUESTIONS ABOUT THE METHODOLOGY OR YOUR PRODUCTION?
Write to us – about this test, your setup, or your next song. We usually get back to you within 3 hours (on weekdays).
- Personal point of contact
- Over 20 years of experience
- Highest quality standards
You can reach us by phone from Monday to Friday from 9 a.m. to 8 p.m.
Silence, sine wave and sweep: Cable sound measured in units of measurement
| Measured variable | VBS L / R | RME L / R | difference |
|---|---|---|---|
| Silence, 20 Hz–20 kHz, unweighted | -113,619 / -113,681 dBFS | -113,516 / -113,565 dBFS | +0,104 / +0,116 dB |
| 1 kHz fundamental frequency (RMS) | -4,20712 / -4,18519 dBFS | -4,20710 / -4,18518 dBFS | +0,000024 / +0,000006 dB |
| Total harmonic distortion H2–H10 | 0,000299 / 0,000231% | 0,000301 / 0,000231% | |
| THD including noise (THD+N) | 0,000584 / 0,000544% | 0,000576 / 0,000540% | |
| Sweep transmission at 1 kHz | -0,19737 / -0,17544 dB | -0,19727 / -0,17535 dB | +0,000103 / +0,000094 dB |
| Maximum deviation across 241 frequency support points | <0,000163 dB | ||
Let's first look at the frequency response: The curves overlap so closely across the entire audible range that they lie on top of each other in the diagram. The difference lies in... less than two ten-thousandths of a decibel – several orders of magnitude below what is discussed in practice.
One striking feature: noise level in silence.
However, the reference condition in the 20 Hz to 20 kHz band is approximately 0,10 to 0,12 dB higher. This is the only value in this measurement series that is noticeably above the noise floor. Before anything is done with this information:
- There are only one still image per conditionNo repetition, no indication of variance, therefore no basis for attributing causes.
- Both cables were replaced together. Even if the difference is real, it says nothing about it. which of the two compounds that caused it – or whether either of them caused it.
- The state of clocking, internal signal processing, and input gain was not completely independently reread after the change.
Furthermore, 0,1 dB in the noise band of a system that has noise at −113 dBFS is far below what matters when music is playing.
Regarding the number of decimal places: Many of the values above have five or six decimal places. That's the Computational solution of the procedureNot the measurement accuracy. We have not determined a calibrated measurement uncertainty for this chain. Anyone who makes a physical statement out of 0,000103 dB is overstretching the data.
Counter-test to the original: does cable sound affect loyalty?
Finally, we directly compared both conditions to the original digital file – that is, to what near the entire analog loop in the project was involved.
| Waveform spacing from the original | VBS L / R | RME L / R |
|---|---|---|
| at unchanged recording level | -40,26948 / -40,52382 dBFS | -40,26950 / -40,52380 dBFS |
| after pure level adjustment (diagnosis) | -41,07187 / -41,12496 dBFS | -41,07185 / -41,12497 dBFS |
Why there is no stable winner
Without level adjustment, the following applies: The reference condition on the left is minimally closer to the original, the test condition on the right. After simple level adjustment The order is reversed in both channels.The differences are approximately 0,000013 and 0,000023 dB respectively – and are therefore smaller than the variation shown by the same condition over its five runs (approximately 0,00017 to 0,00025 dB).
Consequently There is no clear winner here. Which condition is numerically closer to the original depends on which channel and level treatment are considered. This is not a result that can be unilaterally reversed.
Two numbers that are often confused
Furthermore, another classification that is often confused: The distance to the original is approximately -40 dBFS, the difference between the two recordings is approximately -110 dBFSThese are two completely different quantities. The -40 dB includes the entire common system transmission – level, phase, filter, conversion in both directions. That is no distortion from a cable, but rather what an analog loop simply does to a signal. Both states demonstrate this equally.
The blindfolded round: 11 to 9
However, measured values don't answer the question that most people are actually asking. That's why we also listened – in a blinded manner.
Structure of the blind test
A test package was built from the recordings for this purpose: 20 fixed rounds, each with three files bearing the neutral identifiers A, B, and X. A and B represent the two states of the same recording pair, X is a byte-identical copy from A or B. Five pairs of recordings, each presented four times, order randomly shuffled, the assignment of A/B and X drawn independently each round.
Furthermore, the following applies to all 60 files: 50 seconds, stereo, 44,1 kHz, 32-bit float, same size, same format, metadata removed, modification time standardized. No level normalization. The assignment key was predefined using a checksum and was not visible until after the evaluation; the definition was subsequently verified unchanged, as were the checksums of all 60 files. Monitoring was not conducted using comparison software, but rather by neutrally switching tracks in a separate test project.
Preference or discernible sound? What was really asked about cable sound?
This report needs to be more honest than most cable tests online.
In all 20 rounds, participants were asked about their preference: "What is better, clearer, more resolving overall?" It was not The question asked: "Is X equal to A or equal to B?"
However, that is a crucial difference. The classic ABX question is a Identification task – it measures whether someone can reliably distinguish between two signals. What we have here is a Preference survey – it measures what someone prefers. Both are legitimate, but you mustn't sell one as the other.
For the sake of completeness: In an early analysis, a statement from Round 1 was incorrectly interpreted as a formal ABX response. This has been corrected. The original recording remains unchanged in the database; the outdated interpretation is marked as such.
The result of the cable sound test
My own impression during the run-through, from the voice memo:
“Basically, the decision was incredibly difficult, but I always have the feeling that with one signal, the voice or the midrange was generally more pronounced or boosted, and generally a bit clearer in context. Whether that's just my imagination, I obviously can't say for sure before we have the test results. […] The signals are fundamentally very, very similar in their sound characteristics.”
By the way: Whether your own studio knowledge is really as confident as it feels can be tested even without a listening comparison – our Knowledge quiz about mixing and mastering It asks questions from everyday studio life and proves more difficult than expected for most.
This is a personal impression from listening – and it's written here exactly as it was heard. It is no measurement resultAnd it is not reinterpreted as one here. When it clashes with the numbers, neither the numbers nor the impression gain: Then both stand side by side, just as they do now.
What the statistics say – and what they don't –
Therefore, to evaluate the results, one asks a simple question: If there were no difference in the evaluation and every choice was pure chance – how often would a result emerge that is at least as skewed as 11:9?
This can be calculated precisely. The null model is a fair 50:50 allocation – the blind draw produces exactly that. The event of interest is a deviation of 10:10. in any direction, so at most 9 or at least 11 choices for one condition. Two pages because no direction was predetermined – we didn't predict which cable would win.
The result: p = 0,824.
To put it simply: By pure chance, an outcome at least as unbalanced would occur in approximately 82 out of 100 such series. 11:9 is therefore a completely unremarkable random result. Additional measurements were taken for verification. one million series simulatedOf these, 823.938 were at least as unbalanced, i.e., 82,3938% compared to exactly 82,3803% – with a simulation error of approximately 0,038 percentage points. The simulation confirms the calculation; it generates no additional audio material and does not increase the sample size. The calculations were performed using the exact binomial test from SciPy, cross-checked with an independent calculation.
Four misinterpretations to avoid when dealing with p-values
Experience shows that this value is regularly reported incorrectly. It explicitly does not mean:
- Not "The cables sound the same with 82% probability." A p-value is not a probability for a hypothesis.
- Not "82% of the answers were guesses." The value says nothing about individual answers.
- Not "55% success rate". There were no right or wrong answers because it was not an identification task.
- Not "This proves that there is no difference." Non-significance is not proof of equality.
And the point that is most often overlooked: Two audibly different versions can still be preferred equally often. An 11:9 ratio is compatible with indistinguishable signals – but equally so with distinguishable signals without a stable preference. Both possibilities remain open after this pass.
The built-in counter-check: 13 out of 20 rounds contradict themselves.
In each round, there were three files, but only two different recordings. X was always a byte-identical copy of A or B. Therefore, in each round there was one twin pair of two identical files and one single file that was different from the other.
This results in a strict control that requires no additional measurement: Declaring one file from the pair of identical files the winner is essentially declaring that file a superior version of its own duplicate. This isn't a mishearing or an incorrect answer – in a preference query, there were no correct answers. However, it is a logical inconsistency, and it can be precisely identified retrospectively using the file checksums.
Result: In 13 of the 20 rounds, the choice fell on a file that had an identical duplicate in the same round. Only 7 rounds were consistent, meaning they were attributable to the single differing file.
Important, so that nothing gets mixed up here: The result of the hearing test remains 11:9. This 13:7 is not a second count of the same comparison, but a different question posed to the same 20 voters. The 11:9 counts which Cable condition was preferred – for that, the answer key is needed. The 13 to 7 counts as whether the selected file was a Duplicate or whether it was unique – the checksums are sufficient to determine that. Every choice carries both labels simultaneously. The fact that the numbers look similar is a coincidence: the expected probability of a random match is 10 out of 20 for the cable comparison and 13,33 out of 20 for the duplicate question.
And now the crucial classification: If someone randomly clicks on one of three labels, they will pick the twin pair in two out of three cases. The expected probability of random selection is therefore 13,33 out of 20. Thirteen were observed. An exact two-sided binomial test against this expectation yields... p = 1,00 The elections are indistinguishable from pure guessing. A Monte Carlo analysis of one million series confirms this.
In other words, the responses contain no information about which two files were the same and which were different. Anyone who wanted to hear the difference would have to name the differing file more often than would be expected by chance, or report a tie. Neither of these things happened.
Was dieser Befund offenlässt
Two explanations remain open, and we cannot separate them. Either one's perception of such similar signals varies by precisely this amount – in which case the described "clearer, more detailed" self-production is the issue. Or the playback conditions of the individual tracks were not exactly identical; this was not observed during listening. Regarding the cable question, both explanations lead to the same conclusion: Based on this, no audible superiority of either condition can be established.
This evaluation was created retrospectively and was not predetermined. It is therefore exploratory – but it is based on checksums, not interpretation, and the assignment of cable conditions to A and B remains unaffected.
The limitations of this cable sound listening test
- A listener. No statement about other ears.
- A musical excerpt, the same in all 20 rounds. 50 seconds from a production. Within these 50 seconds, however, the bandwidth is considerable: The short-term level fluctuates by around 26 dB, with a difference of 18 dB between the 10th and 90th percentiles. The crest factor ranges from 7 to 20 dB, and the spectral center point from 643 Hz to almost 8 kHz – from quiet and thin to loud and dense. A level- or density-dependent effect would have had the opportunity to manifest itself here. However, the result does not automatically apply to other productions, mixes, and mastering stages.
- More music excerpts would not automatically have been better. Analyzing additional material without predefining the analysis primarily increases the likelihood of a lucky guess. More material is only helpful with a predefined analysis and appropriate sampling plan – otherwise, it's more likely to lead to a seemingly significant result than a reliable one.
- Five pairs of recordings, listened to four times. These are repetitions of the same recordings, no 20 independent hardware changes.
- Both cables were replaced together. No isolation of individual connections.
- Not a standards-compliant ABX or ITU test. We do not claim conformity to any standard; the Methodological framework of ITU-R BS.1116-3 This is only for contextual purposes.
- The playback settings were not independently logged during listening. Import, fader, pan, effects, and output assignment are not independently documented. A subsequent review of the project cannot retrospectively prove its original state.
- Round 1 as an isolated case: There, the choice fell on B, while A and X were described as indistinguishable – even though B and X were identical files. This round is one of the 13 contradictions counted above.
Cable sound in conclusion: what that means – and what it doesn't.
What we can say about cable sound
In summary: Under the conditions documented here – this studio, these devices, this piece of music, this listener – both states delivered virtually identical signal transmission. In the tested digital excerpts, the data arrived flawlessly at the 24-bit level. The blinded preference test showed no systematic advantage. And no consistent ranking emerged compared to the digital original.
What we cannot say
- That audiophile USB cables cannot have any influence whatsoever. This test examines two specific cables in a specific chain – not the principle.
- That the differences are inaudible. This would require an identification task, which we did not perform.
- The slight noise variation during silence is due to the cable. However, one silent recording per state, with both connections switched simultaneously, does not establish causality.
- The manufacturer's claims regarding microstructure, geometry, and time behavior are incorrect. They are simply not compatible with this design. Not checked.
Fair to the manufacturer
Andrej Sokić, the manufacturer, gave his prior written consent to the entire procedure – blinded hearing test, objective measurements, long-term observation – even though it was clear that we would also publish a negative result. He provided the cables free of charge and approved the publication without influencing the outcome. This is more transparency than is customary in this industry, and it belongs in this report just as much as the figures.
Fair in relation to the question
Conversely, if we truly wanted to answer the question "Can the states be reliably distinguished by sound?", we would need a separate identification series with a predefined evaluation and a sampling plan tailored to the effect we were trying to find. We deliberately omitted this – one doesn't repeat a test until the desired result appears.
What I personally take away from this
Two things.
First: The effort was definitely worthwhile – not because of the result, but because of the method. The repeat comparisons were the most important part of this measurement series. Without them, I would have seen a residual difference of −109,6 dBFS between the states and probably thought that was significant. Only the measure of "how much does the same system fluctuate with itself?" makes the number meaningful. Anyone who publishes such tests without this step is publishing their own noise.
We offer precisely this principle – check first, then decide – as a service: In the case of a mix analysis We listen to a production in detail and identify where its real potential lies. And anyone who wants to know what our mastering will do to a specific song before an entire album is produced can find out with a single song as a test mastering Find out – a verifiable result instead of a promise.
Why blind listening is uncomfortable
Secondly: During the listening test, I thought I could hear something – clearer midrange, a bit more resolution. After the blinding was lifted, the ratio was 11:9. This is exactly the kind of experience you have to have yourself to understand why blind listening is so uncomfortable and so necessary when there are only small differences.
Ultimately, what remains is a simple recommendation for anyone facing the same question: Before investing four-figure sums in a connection, it's worth taking a look at where the real leverage points lie within your own setup. Based on that, what a complete mixing and mastering production costsThe price of a single cable is already disproportionate to its demonstrable benefit. Room acoustics, listening environment, and the quality of the recording itself In our daily work, we deal with magnitudes that we are discussing here in ten-thousandths of a decibel. Even a cleanly configured audio computer It provides more stability in everyday use than any cable.
Methods and data appendix for cable sound measurement
Finally, this section is for anyone who wants to recalculate or replicate the project. Those who only wanted the result can find it above.
Recording and music comparison
Format: 44,1 kHz, stereo, 32-bit float. Music source: 2.675.375 samples (60,666099773 s). Raw music file: 317,875873 s or 320,259524 s, five complete passes extracted each – integer trimming only, no processing.
Music comparison: Identical inner section 5–55 s. Error level = RMS of the direct waveform difference. No gain, tone, fractional sample, or resampling correction. 25 cross-comparisons, 10 repetition comparisons per condition.
Test signals: Silence, sine wave and sweep
Silence: common inner window 1–28,889 s. Welch method, 2-s Hann window, 50% overlap, mean distance per window, band 20 Hz–20 kHz, unweighted (no A rating).
1 kHz sine wave: Identical file window 2–5 s. Frequency adjustment of the fundamental tone, least-squares fit, harmonics H2 to H10. Total harmonic distortion including noise from the fundamental-corrected remainder in the band 20 Hz–20 kHz. A method check with a synthetic signal of known distortion (0,1% second harmonic) yielded −60,000005 dBc compared to a target value of −60 dBc.
sweep: Source 34 s total (2 s silence + 30 s logarithmic sweep 20 Hz–20 kHz + 2 s silence), peak level −18 dBFS, fade-in/fade-out times 50 ms and 20 ms respectively, both channels identical. Both feedback loops referenced to the same complete source. Identical FFT processing, quotient of summed spectral powers in 1/24-octave bands, 241 support points, absolute and additionally referenced to 1 kHz per channel. A synthetic control with a known gain difference of −6 dB was passed. The edge bands are interpretable to a limited extent due to fade-in/fade-out times and sweep limits; a harmonic-resolved distortion measurement across the frequency range was therefore not performed.
Phase and group delay in cable sound (Addendum 21.09.2026): From the same sweep recordings, read-only and without a new measurement run. Both channels individually: integer alignment via cross-correlation, then the transfer function of one condition against the other as an FFT quotient. Only points at least 60 dB away from the spectral maximum in the 20 Hz to 20 kHz band were evaluated, resulting in 950.138 points per channel. The phase was deconstructed and the linear component separated as a constant residual delay; the group delay difference per band is derived from the slope of the phase versus frequency.
Spectrograms and why difference images can be misleading
Procedure: Short-time Fourier transform with symmetrical Hann window 4096, step size 2048, no mean removal, one-sided power density in dBFS/Hz. Main scale −165 to −15 dBFS/Hz; noise and residual signal views separately −175 to −115. Overlays with common scale, no individual normalization. Welch comparisons: Hann 16384, at silence 65536, 50% overlap. Spectrogram definition from SciPy is the basis of the presentation.
Why difference images should be read with caution: A difference map of the form 10·log₁₀(power density A / power density B) is no spectrogram of a waveform differenceRandom noise is never identical in every frequency band; in areas with very low absolute levels, this results in large relative deviations that appear colorful but are meaningless. Therefore, this article includes the absolute scale and this explanation with every such graph.
What the test doesn't say about cable sound
All values mentioned in the text are taken from the archived evaluation files of the measurement runs and were checked against these files and their respective evaluation windows for the purposes of this article. A complete source and statement register, which traces each figure back to the file, window, and procedure, is available in the project archive.
What cannot be determined from this data: Crosstalk between the channels (the excitation was identical on both channels), calibrated jitter, isolated electrical cable characteristics, and any statement about devices, rooms, music, or listeners outside of this setup.
Complete chain: devices, cabling, power supply
Added on September 21, 2026, at the manufacturer's request. All information remained unchanged. beide Cable conditions – nothing was changed on this chain for the cable sound comparison.
- Computer and software: Windows DAW computer, Cubase 14 Pro, 44,1 kHz, 32-bit float.
- Driver: RME MADIface-USB 1.0.17.0 for ADI-2 Pro and Fireface UFX+.
- Converter in the measurement path: RME ADI-2 Pro FS R Black Edition, reference level +19 dBu, internal clock source, own external 12-volt power supply – no power supply via USB.
- Interface in the eavesdropping path: RME Fireface UFX+, clock source AES, buffer 1024 samples.
- Patchbay: Flock Audio Patch, path 23 left and 24 right, direct without any looped devices.
- Cabling: The entire studio cabling was planned and installed by ES Pro Audio (Erwin Strich). Analog and digital connections are consistently Vovox and Sommer Cable. The AES3 connections to the first monitor and from its THRU to the second monitor are Vovox, terminated to length without any coiled excess cable.
- monitors: Dutch & Dutch 8C, AES3 via 110-ohm XLR; crossover, correction and D/A conversion take place in the speaker.
- Power supply: Standard power strips, no power conditioning or filtering. All devices are running with their original cables; no power cables have been replaced.
- Measuring instruments: Measurements are taken using the studio's own signal chain. There is no separate test setup with its own power supply because the same converter stage is used for both measurements.
Individual part numbers for the cable types are not documented and are not stated here. What's crucial for the cable sound comparison is something else entirely: nothing in this chain was changed between the two cable conditions.
Conflict of interest and data holdings
The cables tested were provided free of charge as product samples and remain with Peak-Studios. No money changed hands, there were no content specifications, and the manufacturer did not approve this text. The manufacturer agreed to the procedure and the publication of a null result in advance.
Raw recordings, blind test packages, answer keys, and evaluation scripts are fully archived and protected against alteration via checksum manifests. They are not publicly available due to copyright restrictions on the test music and test integrity. We are happy to answer any questions regarding the methodology directly.
Frequently asked questions about cable sound
Can audiophile USB cables actually affect the sound?
Not during the data transmission itself – that is error-protected. Either a data packet arrives correctly, or it doesn't arrive at all. Therefore, discussions usually focus on indirect causes: grounding, shielding, and interference in the analog environment. Our test examined both: data transmission (error-free at the 24-bit level) and analog output (virtually identical). However, it only answers the question for these two cables in this chain.
Is there such a thing as cable sound with digital audio cables?
In our series of measurements, no cable sound was detectable with USB. This cannot be answered in general terms because "digital audio cables" refers to four different transmission methods. With USB in asynchronous mode, the device itself provides the clock signal – any influence from external sources is technically difficult to explain and was not detectable in our measurements. With AES3, coaxial S/PDIF, and Toslink, the clock signal is embedded in the signal; here, impedance errors and reflections can indeed have measurable effects. Our test covers only the USB connection.
How can I tell if a digital audio cable is truly defective?
The issue lies in dropouts, crackling, and increasing error counters – not in a change in timbre. In addition, there are connection drops. Many audio interfaces display the driver's error counters directly. If this counter remains at zero, the connection is working flawlessly in the tested area.
Why not a classic ABX test?
Because this round asked about preference, not identification. To subsequently market this as ABX would be dishonest. A reliable statement regarding audible differentiation would require a dedicated series with a pre-defined evaluation and appropriate sampling plan.
Questions about test design and validity
Does 11 to 9 mean that the expensive cable won easily?
No – with 20 random decisions, 11:9 is a completely unremarkable result. It occurs purely by chance in approximately 82 out of 100 cases, at least to this extent. No advantage can be derived from a ratio of 11:9.
Why were both cables replaced at the same time?
For practical reasons during ongoing studio operations – and because the realistic comparison is between the delivery state and the test state. The price for this is that the cause of any difference cannot be attributed to a single connection. This limitation is therefore stated at every relevant point in the text.
Is the result transferable to other studios?
Only to a limited extent – measurements were taken on a specific chain with specific devices. Different interfaces, mass ratios, and spaces may behave differently. The method is transferable – the result is only a guide.
How reliable are measurements with five decimal places?
They represent the computational resolution, not the measurement accuracy. We have not determined a calibrated measurement uncertainty for this chain. Therefore, we only interpret differences if they are larger than the observed variance for the same condition.
Technical terms and practice
What does dBFS mean?
dBFS is the digital level relative to full output: 0 dBFS is the maximum, all values below it are negative. The abbreviation stands for "decibels relative to full scale". A residual signal of -110 dBFS is therefore extremely quiet – around 110 decibels below full scale.
What is a null test and why is it insufficient here?
In the null test, two signals are subtracted from each other; if exactly silence remains, they are identical. This works for purely digital copies. However, as soon as a signal undergoes digital-to-analog and analog-to-digital conversion, some residue always remains – converter noise and minimal fluctuations. Therefore, a comparison of the condition with itself is also necessary as a benchmark.
What should you focus on instead if you have a budget?
Room acoustics, listening situation and the quality of the recording itself – in that order. This is based on our daily experience. These three points involve orders of magnitude that are debated in cable discussions in ten-thousandths of a decibel.
