Using Statistics to Overcome Fit Variation for Headset Measurements (100 Things #81)

SoundCheck’s statistics allows for multiple measurements from a sequence to be analyzed together, to determine results like average and standard deviation. Using statistics can overcome placement variations when measuring headsets on a head and torso simulator. This sequence demonstrates a measurement of a USB headset, performed five times, where the headset is removed and repositioned on the HATS each time. SoundCheck’s statistics then take these five measurements, account for the differences in placement, and display an accurate set of measurement results. If you are testing to standards then statistics makes measurement of communications devices fast and repeatable.

Using Statistics to Overcome Fit Variation for Headset Measurements

Try measurement statistics in SoundCheck

Learn more about our pre-written TIA-920-B test sequence mentioned in this video. This pre-written sequence tests to TIA 920-B, a comprehensive US dual-bandwidth standard that applies to both narrowband (NB) and wideband (WB) devices.

Video Script:

Statistics have been a feature in SoundCheck for a very long time. But did you know that statistics can be used to overcome fit variation when measuring body-worn devices? Let’s look at how we can use this feature to make repeatable TIA-920B measurements on headsets?

For realistic and accurate results, headsets and other body-worn devices should be measured on a Head And Torso Simulator, or HATS, placed just as worn by real users. Unfortunately, small changes in position can lead to significant changes in both the level and the sound quality, whether on a real person or HATS.

For example, when placed carefully, the receiver of our USB headset sounds like this. (Audio example 1: proper placement)

When placed poorly, it sounds like this. (Audio example 2: improper placement)

To obtain repeatable results, we make several measurements and average the results, using the Statistics Step.

The headset is completely removed from HATS after each measurement, then repositioned for the next. With practice, 5 measurements are usually enough. This procedure is defined in ITU-T P.380 and IEEE 269 and used in Listen’s pre-written sequences that implement TIA 920-B.

Let’s make some measurements.

We are testing a USB headset that has two receivers and a boom microphone, intended for speech communication. There may be some speech-sensitive signal processing, so the test uses real speech. The signal is played out to the receivers first, then to the HATS mouth.

When the first measurement is finished, the receive frequency responses and single parameters such as output level are shown on the top line

In a similar way, the sidetone frequency responses are on the second line, and the send frequency response is on the bottom line.

After 5 measurements, with re-positioning between each measurement, we can see the individual frequency responses. Let’s take a closer look at the Left receive frequency response. The individual frequency responses are in gray, the current measurement in white, and the current mean in blue.

After the last of the measurements, we can see the mean results. Tolerances from the standard have been applied to the mean receive and mean send frequency responses.

The standard deviations show the repeatability of the individual measurements. If the standard deviations are within the tolerances, the mean results are acceptable. When results from 2 or more operators using this method are compared, the mean results will usually be very close, even if the individual measurements are somewhat different.

Statistics helps overcome fit variation to make accurate and repeatable measurements of headsets, as well as most other body-worn devices such as helmets, goggles, parrots and so on. And, if you are making TIA-920-B measurements on such devices, you can save a lot of test writing time with our pre-written TIA-920-B sequences. These can be used for USB, Bluetooth or wired analog devices, and there are also open-loop sequences for testing devices that connect to a server.