2025-07-01 Update: We are now using a different method for testing and visualization of audio products and results. This article supports the testing done in the ShortCircuit videos linked below.

You may have seen some audio graphs in a recent ShortCircuit video or two and wondered how we collected those measurements, or - depending on your familiarity - even how to read them. This article is an introduction to the tests that we've debuted in those videos, and a bit of analysis of the new WH-1000XM6. First we'll cover the test equipment that we have set up, and then we'll get into the test results and graphs.

We will be using the results of the Sony WH-1000XM5 and WH-1000XM6 as examples for the purpose of this article.

Test Setup

Our testing chamber is a 3.00 m x 2.40 m x 2.40 m(9'10" x 7'10" x 7'10" for Americans) standalone room treated with 4-inch NetWell MelaMax open-cell melamine anechoic foam panels. This is with the aim of creating a semi-anechoic environment by minimizing internal reflections and external noise, but we are still working on eliminating some of the effects of the outside world. It is quiet in the room, but not quite to the point of driving you crazy.

While a listening room or other more realistic environment can also provide some value when audio testing, the consistency of a quiet and predictable space will allow us to eliminate as many variables as possible. While a speaker's interaction with their environment is important, most headphones aim to completely isolate their listener from the environment. When testing the isolation(including active noise cancelling(ANC)) and microphone frequency response we are able to add in noise at predictable levels to simulate a louder space.

Inside of the audio testing room we use the following equipment in the following arrangement:

• Head and Torso Simulator - B&K 5128 HATS

• Leakage Microphones - Dayton Audio EMM-6 (x2)

• Audio Interface - Focusrite Scarlett 18i20 (3rd Gen)

• Sound Calibrator - B&K 4231

• SPL Meter - Galaxy Audio CM-140

• Speaker - Mackie MR824 (x2)

• Speaker - Mackie MR524 (x2)

• Subwoofer - Mackie MRS10

The star of the show is the B&K 5128 HATS(Head and Torso Simulator) which is notably just a HS(head simulator) in our setup. This is fashioned after a human head with 'realistic' ears and ear canals meant to simulate the interactions of sound waves with a real head instead of just placing microphones inside of the microphone ear cups. It also features a speaker - where a mouth would be - to enable microphone testing.

We use the four Mackie speakers in a 4.1 setup to simulate environmental noise through pink noise and other audio samples. The EMM-6s are positioned 15cm away from the ear on either side of the head for measuring audio leakage.

Prior to testing, we explore the documentation and existing literature on the product to find any notable features that will require extra attention or circumventing. During this time we also just listen to the headphones to begin forming our subjective evaluation. We update all software/firmware involved and take note of software versions, EQ, filters, settings, and other factors that may affect the final result.

Tests

Frequency Response

The first and most fundamental test that we perform is the frequency response measurement. It visualizes how a headphone reproduces audio in the 20 to 20 kHz frequency range. This test is the basis for the "Frequency Response - Raw", "Frequency Response - Compensated", and "Frequency Response - Consistency" graphs. We use logarithmic sine sweeps and full-range pink noise stimuli played through the headphones, measuring through the HATS with the headphones in five different positions on/over the ear.

These graphs also include our LMG Headphones Target Response for the B&K 5128, which represents the average response that most listeners will perceive as balanced. Notably, this is not a 'flat' frequency response, and that is because a truly flat response will sound dull to most listeners. The target frequency response is a recreation of what listeners find enjoyable and sounds neutral. It is also important to note that we reposition the headphones multiple times to account for fit differences. The frequency response at high(above 10kHz) and low(below 100Hz) frequencies can differ quite significantly from person to person.

Frequency Response - Raw

This is the most basic of the frequency response plots, it is the raw data that was measured by the HATS eardrum microphones. It is not particularly useful for a general audience as it can be cryptic to read but it can be useful for engineers and enthusiasts.

The basic principle of the frequency response graphs applies here, frequency ranges with measurements above the target may sound overemphasized while frequency ranges with measurements below the target may seem to be lacking. For the WH-1000XM6(note we only measured the right ear cup) we can see that there is an overemphasis in the "Sub bass" and "Bass" range, though this is slightly easier to see in the "Frequency Response - Compensated graph". Some users will actually prefer this type of bass-heavy sound.

Frequency Response - Compensated

Likely the most useful of the frequency response graphs is the compensated graph. This is created by subtracting the headphone target from the measured frequency response. This produces the difference between the measurement and the target, essentially the "error". From this we can identify deviations in wide bands as well as more narrow ones like the one that exists near 2.2kHz for the WH-1000XM6.

This is the most useful as we can see precisely where the headphones are prone to overemphasis. Instead of comparing two squiggly lines on a graph, the target is a horizontal line at 0dB and any deviation above or below the target shows the difference. For the WH-1000XM6, we can more clearly see the up to 5dB overemphasis in the Sub bass and Bass regions. The deviations above 10kHz are slightly concerning, but as previously mentioned, the sound reproduction of the headphones can vary depending on fit and placement.

Frequency Response - Consistency

The next visualization of the frequency response highlights the consistency. This intends to show how different the headphones may sound to different people, or even the same person depending how it is placed on their head. We have five 'standard' placements of headphones on people's heads: centered, forward, back, up, and down. This way we can position the headphone drivers differently with respect to the ear and potentially also determine if the padding around the ear cup will cause inconsistencies. These measurements are all displayed as a difference to the average of the measurements.

For our example WH-1000XM6, we see that the frequency response is very consistent for frequencies below 500 Hz and then it begins to deviate depending on the fit, worsening in the 8 kHz to 20 kHz range. This is a relatively common pattern as most ANC headphones will have microphones inside that will measure the response and compensate low frequencies for fit and seal differences. Higher frequencies are 'above the reach' of compensation and they are often highly dependent on the topology of the ear, resulting in more variation.

Noise Isolation

This is one of the more complicated tests, but also arguably the most interesting results. We play calibrated pink noise at 90dB(A) SPL from four speakers directed at the HATS and take measurements with no headphones, with the headphones on the HATS but ANC deactivated, and with the headphones on the HATS with ANC activated. We also repeat this process for five headphone positions to determine the fit variance. The measurements with no headphones is then defined as the "Target"/reference and the deviation from this reference is plotted with ANC both activated and deactivated.

In some cases you may find the measurements with ANC deactivated referred to as "isolation", this is just the physical isolation that the headphones provide by blocking external sounds. This effect is often more significant in the higher frequency ranges as you may know intuitively from hearing only the bass component of your neighbour's loud music.

For the WH-1000XM6 we see a fairly standard 20 dBr drop for frequencies below 200-300 Hz, but it also extends some additional noise cancelation up to 1 kHz.

Microphone

The microphone frequency response is another interesting test to conduct because it uses the mouthpiece speaker of the HATS to play recorded speech while the speakers on the perimeter of the room can produce pink noise or other sound effects. The frequency response graph shown below is with no background audio added in so that we can see the frequency response of the microphone itself. Then the external audio sources can be added in to simulate a less ideal situation.

Below is the microphone frequency response of the WH-1000XM6 along with three sample microphone recordings. The frequency response graph can be more useful for proper microphones but the sample audio is likely the most useful if you know you're going to be taking a lot of phone calls or communicating(trashtalking) with teammates(casuals) online. Here we can see and hear that the WH-1000XM6 suffers by not having a boom microphone, but is able to capture frequencies in the range of 300 - 7,000 Hz, generally an acceptable range for capturing the human voice.

As always, let us know what you think or if you have any questions about any of the tests or results above! Feel free to use our Google Form or leave your thoughts as a comment on the ShortCircuit video. We've rolled out the tests above to get some feedback and we'll likely spend some time implementing feedback that we gain!