Conclusive test: a simple way to test audio components by ear

With conclusive tests we can quickly check whether our system is free from certain performance issues, not just trivial things like noise. They come handy when measurements are not available for a product, or in controversial cases when an unexpected positive result is caused by a hidden parameter.

August 23, 2026

Testing audio components with music is not only a tedious process, but we can hardly separate and identify different artifacts, such as frequency response errors from nonlinear distortion. However, certain artifacts (audio fidelity parameters) can be easily tested even without measurements. Testing these characteristics with special signals can be almost as reliable as measurements. Of course, this doesn't mean that we can get rid of measurements (not to mention that the results may change from person to person, due to individual differences).

If measurements are available for your gear and measurements tell that your system is "flawless" (no audible coloration from DACs, amplifiers, some minor coloration from speakers...) then you can skip these tests.

A fundamental misconception

A common mistake audiophiles make is to trust their hearing completely. An opposite - and extreme - view is when someone only accepts measurement data or ABX test results and doesn't trust his or her hearing at all. Although we can't trust our hearing all the time, we can trust our hearing sometimes. Instead of "believing in measurements" (and believing in ABX) we can learn how to identify some possible errors in the audio chain.

Here is another problem. If our claim is that hearing is unreliable and inaccurate, then we are in trouble because accuracy is the basis of science. What we can do is to limit our interest in those cases when our hearing is reliable and build a system from reliable listening tests and signal processing.

What can we detect easily?

Reliable judgement is not only about hearing, but also about focusing. If we can find a signal that allows us to easily hear and easily focus on the difference, then discrimination is simple. And when is focusing easy? When we have no other option to focus on anything but "one thing". Detection of noise, different types of nonlinearities and resonance can be easy because they depend on hearing thresholds (masked, non-masked).

Human hearing is very good at the following:

detecting any type of sound in silence;
detecting noise in the presence of a pure tone or harmonically related tones;
detecting non-harmonically related tone(s) in the presence of another tone (pure tone, two tone, chord with few tones);
detecting high Q resonances in pulses or pulse series (at least 10 msec interval for the pulse series);

Human hearing is also good at detecting pure tones or harmonically related tones in noise, but this has little relevance in component testing.

What is a conclusive test?

Conclusive test method has five requirements:

One interpretation
Result can only be interpreted in one way. There is no other way.
High sensitivity and accuracy
Easy to hear the artifact (usually this is achieved with a signal with which the error is best heard). No tiring comparative tests (Bye-bye ABX).
No reference system is required
We can use these signals without listening to a reference system.
No special training is required
Doesn't require special training or experience.
Fast and "comfortable"
Most of the tests only takes a minutes and don't cause ear fatigue.

The essence of the whole method is that these tests are reliable without ABX testing. The addition of ABX test can be useful in certain controversial cases, but due to the high sensitivity of the method even in these cases we can say that the distortion, noise or ringing can't be heard with music. Besides, a well designed conclusive test is free from false cause type errors (~hidden variable) and correlation between the data means causation.

There are some measurement signals that cannot be used in a conclusive test. For example, testing nonlinear distortion with a pure tone is not a conclusive test because we need a reference system for a proper judgment. Using a pure tone without a reference system we can only easily detect excessive clipping but not moderate distortion.

Conclusive test	Test with music
One interpretation.	Ambiguous.
Reliable, accurate.	Unreliable, inaccurate.
No reference system required.	May require a reference system.
No training required.	May require training.
Fast - no ear fatigue.	Slow - ear fatigue very quickly.

Comparing conclusive test with testing with music

Some test methods are related to only one type of component (digital artifacts are related only to DACs or software), whereas some test methods qualify the whole system and all components when the test is negative (ok), but qualify only the whole system when the test is positive (fail).

And here is the list about the artifacts or "fidelity parameters" that we can easily make a good judgment without measurements:

Noise: any kind of noise in the system (residual noise, self noise, quantization noise...)
Sampling /resampling related distortion.
Nonlinear distortion in the high frequency range and TIM!
Anti-aliasing related distortion in DACs (anti-imaging related distortion actually).
Resonance / phase-shift test with impulses.
Channel balance (center image position).
Crosstalk.

Crosstalk is not discussed in more detail because playing with the balance control is very simple and straightforward (besides, it's hardly a problem with modern electronics). Channel balance is omitted for the same reason (we only need to listen to a mono source).

Some basic parameters that determine audio fidelity can't be tested in a simple way. For example, there is no simple, fast conclusive method for frequency response issues, or a comprehensive testing of nonlinear distortion. However, there are some tests for some types of nonlinearities.

Thanks to the high sensitivity of these tests, slightly positive test result (barely audible distortion) is acceptable in nonlinear distortion tests and anti-imaging tests. Slightly positive test (barely audible distortion) is not acceptable when testing the quality of resampling.

Signal levels

Signal levels are important, especially when high-frequency test tones are played on loudspeakers with small tweeters... We want to "overload" the system, and not a destructive test. If the volume is set to the appropriate level with music, then -20 dBFS above 10 kHz is completely safe for tweeters (1 Watt from 100 Watt). -10 dBFS is safe if duration is limited (in the test files duration is limited to two seconds). With headphones there are no such restrictions. (dBFS: decibel below full scale, 0 dBFS is the level of a full scale sinusoid).

There are other benefits of setting the level of test tones around 20 kHz to -20 dBFS. One benefit is that -20 dBFS is close to the possible peak level (it's unlikely that music level goes above -20 dBFS around 20 kHz). In addition, -20 dBFS corresponds to approx. 85 dBSPL, which is the lowest human threshold at 20 kHz. It's extremely unlikely that one would hear a 20 kHz tone at -20 dBFS.

You can adjust the volume with the following video (use the "master volume control", not the volume control in the video player).

Direct links to audio tests:

Noise floor

Applies to: digital audio players & DACs, amplifiers
Test required in advance: none
Test signal: none, 16-bit dither

Noise is not an issue in modern DACs and amplifiers, but I added to the list because it is of fundamental importance. In addition, investigating noise-related issues is sometimes not easy.

Testing noise in a system is quite simple as no (audible) signal or music is required. Testing noise in DACs and amplifiers with digital input is a bit different from testing noise in amplifiers with analog input. In order to test DACs and amplifiers with digital input we need to play a FLAC or WAV file with 16-bit dither or 24-bit dither, but the self noise of amplifiers with analog input can be evaluated without playing anything. However, if we listen to loudspeakers using an amplifier and a digital audio player, then we can test the two devices at the same time, by listening to the dither. We only need to test them separately if we are not satisfied with the results. However, if the signal transmission from the digital audio player to the amplifier is digital, then any audible noise is generated inside the amplifier (this is true if equalization is disabled in the source).

When listening to the following audio sample you shouldn't hear anything from the speakers or headphone.

Noise test: 16 bit dither

Note: for this test the audio will start over again.

It goes without saying that a noise test requires complete silence. If we are unsure, we can turn the device on and off to determine if we hear a difference or not.

The volume knob trick. By turning down the volume on the amplifier we can make some judgements on the origin of the noise, since the volume control can only affect the noise before the volume control, but not the noise after the volume control. If we have a noisy system and by turning down the volume on the amplifier the noise has gone, then the noise comes from either the preamp stage or from the audio player. If the noise remains after turning down the volume on the amplifier, then we know that noise comes from the power stage. The audio player may also be a "noise source", but at least we know that the amplifier is part of the problem. (In real life it's very unlikely that both has audible noise.).

Resampling test (distortion from resampling)

Applies to: digital audio players & external DACs (primarily software resampling)
Test required in advance: noise (recommended but not necessary)
Test signal: pure tone above 10 kHz (with level between -10 dBFS and -20 dBFS)

Resampling is considered a bad thing that should be avoided at all costs, yet it's hardly a problem for a long time. Even the cheapest Android phones can resample from 44.1 kHz to 48 kHz without audible issues. Testing the quality of resampling is more important when custom software or hardware is used.

Distortion from resampling can be tested quite easily by listening to high frequency pure tones (sine signal). A pure tone test at 20 kHz set to -20 dBFS is enough, but any frequency above 10 kHz works well.

When listening to the following audio samples you shouldn't hear anything from the speakers or headphone. Level is set to -20 dBFS, frequency of the test tone is 20 kHz. If one version has a distorted sound, then poor quality sample rate conversion is used on this version. Silence means lack of resampling or high quality ("transparent") resampling.

Sampling rate: 44.1 kHz / Tone @ 20 kHz

Sampling rate: 48 kHz / Tone @ 20 kHz

We can also test resampling with a high-frequency but audible pure tone. The test signal is a 10 kHz pure tone set to -15 dBFS. The test tone should sound clean, with no distortion or any "second tone" audible. The benefit of this test is that we don't have to worry about hearing thresholds above 16 kHz, as anyone with average hearing can hear this sound.

Sampling rate: 44.1 kHz / Tone @ 10 kHz

Sampling rate: 48 kHz / Tone @ 10 kHz

Why pure tone is the best test signal for testing the quality of resampling filters? Resampling creates frequency components that are harmonically not related to the original signal. Even with a pure tone the distortion has a very dissonant, unpleasant sound. The distortion also increases with frequency. Ideally pure tones at any frequency should sound the same at 44.1 kHz and 48 kHz sampling rates. If high-frequency tones sound different at different sample rates and one version has a distorted sound, then a poor quality sample rate conversion is used on this version. The other sample rate is likely free from software resampling.

The goal is lack of audible distortion from resampling and not bit perfect playback. Knowing that resampling doesn't generate audible distortion in our system is more important than forcing bit perfect playback. Bit perfect playback is just a solution when transparency can't be achieved by other means.

Nonlinear distortion in the high frequency range, TIM

Applies to: digital audio players & DACs, amplifiers, loudspeakers, headphones
Test required in advance: sampling rate conversion distortion
Test signal: high frequency two-tones

There is no simple, fast conclusive method for a comprehensive testing of nonlinear distortion. However, there are some test signals which can reveal a bit more then just listening to pure tones or music. We can also easily test nonlinear distortion in the high-frequency range (even Transient Intermodulation Distortion!), and depending on signal level we can make predictions on mid-band nonlinearity (or static nonlinearity).

An interesting test signal is a two-tone with components at 19 kHz and 20 kHz, level set to a reasonable level (between -20 dBFS and -10 dBFS per component). In an ideal system this signal shouldn't generate any audible sound. However, this signal doesn't reveal anything about the third harmonic distortion.

19 kHz + 20 kHz

In order to move the frequency of the third harmonic distortion in the sensitive part of our hearing range, we have to increase the distance between the two tones. This means that the lower tone will be audible. The trick is to listen to a 13 kHz tone and then listen to a 13 kHz tone mixed with a 20 kHz tone. If they sound the same, then no audible second and third harmonic distortion is generated.

First: 13 kHz, second: 13 kHz + 20 kHz

Playing with high frequency two-tones we can test:

amplifier, DAC high frequency distortion (TIM for amplifiers),
amplifier, DAC static nonlinearity (distortion between 200 Hz and 10 kHz) if signal level is higher than -10 dBFS,
loudspeaker or headphone distortion around the test frequencies.

Playing with high frequency two-tones we can't test:

amplifier, DAC nonlinearity in the low frequency region (below circa 200 Hz),
amplifier, DAC static nonlinearity (distortion between 200 Hz and 10 kHz) if signal level is lower than -10 dBFS,
loudspeaker, headphone nonlinearity outside the test frequencies.

Anti-image filter related distortion in DACs

Applies to: DACs in digital audio players & external DACs
Test required in advance: sampling rate conversion distortion
Test signal: pure tone between F_s/2 and F_s/2 - 1 kHz (with level between -10 dBFS and -20 dBFS)

It's very unlikely that this type of artifact affects the fidelity of normal playback. If your system passes the high-frequency distortion test, you can skip this test. However, the related test method is very simple.

When listening to the following audio sample you shouldn't hear anything from the speakers or headphone. Level is set to -20 dBFS.

Tone @ 21550 Hz (F_s/2 - 500 Hz)

What the hell is anti-imaging related distortion? During the reconstruction of the sampled signal from data points, image frequencies appear above the Nyquist frequency (22.05 kHz for 44.1 kHz sampling rate). The role of the reconstruction filter is to attenuate these image components (the reconstruction filter is actually a resampling filter). If these image frequencies are not attenuated well, they may generate distortion in amplifiers or loudspeakers and some of these distortion products may appear in the audible range. Note that we need both poor filtering and high level of nonlinearity above 20 kHz for this type of distortion.

Resonance / phase-shift test with impulses

Applies to: loudspeakers, headphones, digital and analog filters, equalizers
Test required in advance: none
Test signal: pulse, pulse series

Time-domain behavior can be easily tested with a pulse or pulse series. It's worth adjusting the pulse width according to the frequency of interest. Short pulses are more suitable for testing high-frequency resonances, and long pulses are better for testing low-frequency resonances. In the following test signal, the width of the pulses is doubled.

Resonance test, pulses

Note: the longer the pulse, the more low-frequency energy and less high-frequency energy it has. Any audible resonance can be easily detected with this test signal.

Phase shift is inaudible in audio components: crossover phase shift (up to 8th order), DAC reconstruction filter phase shift can't be detected by the human ear even with pulses. Except for a few extreme cases, phase shift due to speaker diaphragm resonances is also not audible (resonances are detected by changes in timbre). If we look at equalizers, a well-designed equalizer either corrects phase or only introduce inaudible phase shift. For loudspeaker/headphone equalization minimum phase filters are preferred, but linear phase filters can also be used from 500 Hz. (Both minimum phase and linear phase filters' impulse response must meet certain criteria, but the rules for linear phase filters are stricter.)

Room standing waves are an exceptional example of audible phase distortion.

Calculations for testing nonlinear distortion with two-tones

The goal is that the quadratic (2nd order) and cubic (3rd order) difference tone appear below 10 kHz, where it's easy to detect. If we don't want to use sample rates higher then 48 kHz, then testing 3rd order nonlinearity requires a signal with a component in the audible range (below 16 kHz). Second order nonlinearity can be tested easily with inaudible tones near 20 kHz.

The frequency of the cubic and quadratic distortion tone, arising from two primary frequency components, f₁ and f₂ (f₁<f₂) is given by:

f_QDT = f₂ - f₁
f_CDT = 2f₁ - f₂

Calculated cubic and quadratic distortion tone frequencies for some frequency combinations:

Test signal	f_QDT	f_CDT
12 kHz + 20 kHz	8 kHz	4 kHz
13 kHz + 20 kHz	7 kHz	6 kHz
14 kHz + 20 kHz	6 kHz	8 kHz
15 kHz + 20 kHz	5 kHz	10 kHz
19 kHz + 20 kHz	1 kHz	18 kHz
16 kHz + 22 kHz	6 kHz	10 kHz
18 kHz + 22 kHz	4 kHz	14 kHz
20 kHz + 22 kHz	2 kHz	18 kHz

Csaba Horváth