Hardware & Sound Personalization DSP Measurements of the Denon PerL & Nura NuraTrue Earbuds

[jjaskuna]

Denon PerL and Nura NuraTrue Hearables
(Forgot this on first post)

I am sharing IR & frequency response measurements for Nura’s NuraTure and Denon’s PerL earbuds. Denon partnered with Nura to re-release Nura’s technology under the Denon (Masimo) name. These devices combine hardware and software (DSP) and it would be nice to measure both.

Is it possible to measure the DSP embedded in an earbud? If we can measure this, how should we use it to parse the hardware’s performance from the signal processing performance? These are interesting questions I start to address in this thread and look forward to your take on the hardware, dsp, and future measurements required.

My proposed approach involves the following steps:

1. Measure the impulse response (IR) of the device without DSP features enabled to capture the hardware performance.
2. Measure the IR with sound personalization DSP enabled to capture the hardware + DSP performance.
3. Compute the transfer function between the two as representative of the DSP filter system operating in the device.

Before we head down the path to derive a transfer function, let’s start by taking measurements to get at #1 & #2. If there is sufficient evidence to indicate some DSP enabled change, and the hardware is solid, then #3 would make sense to complete.

Measurement Details
All measurements are taken on an APx517B and GRAS 45CC setup in my acoustics lab. While not an anechoic chamber, I can easily measure to the equipment’s noise floor.

I use my system to calibrate BT hearables for use in my hearing test and as a result, my measurements are taken at full scale. Note this difference if you attempt to compare my measurements to others.

My measurements are taken at 1.0 FS & 127 absolute volume over AAC. Microphones are calibrated to 94dB at 500Hz. And, to get technical, my measurements are taken one mile above sea level with the corresponding decrease in atmospheric pressure.

I include 5 different measurements taken by unmounting and remounting the earbuds in the ear simulator. Remounting helps capture the variability we experience when using the devices to assess how sensitive the device is to placement. Both the PerL and NuraTrue are extremely sensitive to placement.

Of note, I had to reduce the measurement window when testing the PerL due to IR tail noise, which I comment on at the end.

Reference Device
I am a big fan of the Blessing 2 IEMs and have measured a pair for comparison. These were powered by the Khadas TEA over AAC BT to ensure all measurements pass through the same BT codec. For a detailed description of the Blessing2 check out Amir’s review HERE.

Let’s look at the acoustic response and impulse response of our reference device.

Reference Acoustic Response Reference—Moondrop Blessing2

Reference Impulse Response Reference—Moondrop Blessing2

Key things to note in the reference measurements are the high frequency extension through 20kHz at >= 95 dB and the tight clustering of measurements. Also, the IR and window needed for the IR is important.

Hardware Measurements
With a reference in mind, we can move on to the measurements for the NuraTrue and PerL. The first IR & frequency response measurements are taken without DSP features enabled, including sound personalization & ANC. I will move through the measurements quickly and summarize at the end.

Acoustic Response—Nura NuraTrue Without DSP

Now for the PerL.

Acoustic Response—Denon PerL Without DSP

Impulse Response—Denon PerL Without DSP

Note my modified window and compare it to the Blessing2 reference in which the full IR is captured. Capturing the PerL’s full IR, including its tail, creates problems that I will show at the end.

Software/DSP Measurements
The second set of measurements are taken with DSP sound personalization enabled. Both devices use the Masimo hearing test, which was completed using a GRAS 45 CC ear simulator for both devices. Another words, the same hearing profile is measured and loaded in both hearables. Here are the measurements:

Acoustic Response—Nura NuraTrue with Personalization DSP Active

Impulse Response—Nura NuraTrue with Personalization DSP Active

Note that a wide window capturing the IR and tail was used for the NuraTrue without issue.

And here is the PerL with sound personalization DSP enabled. Note, measurement #5 had a fitting issue in channel 2 and was replaced by measurement #6.
Acoustic Response—Denon PerL with Personalization DSP Active

Impulse Response—Denon PerL with Personalization DSP Active

Summary Plots
The following image plots both devices with and without DSP against the reference device.

Acoustic Response—PerL vs NuraTrue vs Blessing2

To make it easier to see DSP effects I have isolated the PerL in the following plot:

Acoustic Response—PerL without DSP vs PerL with Personalization DSP Active

Closing
Denon’s product modifications do alter the frequency response compared to NuraTrue, largely in the 1- 10 kHz. Further, the DSP enabled response is smoother and closer to the reference response, which is shown by plotting the personalized response of both devices against Blessing2. The degree of improvement is up to your interpretation.

Acoustic Response—PerL without DSP vs PerL with Personalization DSP Active vs Blessing2

Of all the PerL measurements, I found the IR to be the most interesting. It changes with the DSP active, reducing the tail’s noise. I had expected the IR to degrade with DSP active and was surprised by the results.


Windowing and the True Response
During the measurement process I first attempted to use my standard window centered to the IR, which led to a very messy response. Here was the first measurement of the PerL with my standard window length & approach:

Acoustic Response Using a Standard Window—PerL without DSP

Do you think it is appropriate to modify the measurement window to intentionally exclude tail noise?

To answer my own question, the full IR, including tail noise, should be included in analysis. When you do this and assess the transfer function with DSP enabled, you start to see that phase changes are affecting the output signal. Looking into the group delay of the system is also informative.

How to interpret these deserve a dedicated thread, as this is long enough for now. Spoiler alert: the IR approach I have proposed was not feasible with this quality of IRs. Other options are available…should the community want to explore these.


For Future Consideration
Here are three graphs to prime the option for further discussion on transfer functions and how we can use them to start measuring DSP features popping up more often. Note, AP allows transfer function measurements using complex and magnitude only derivation. This is one of many approaches. Learn more about TF measurements with AP’s TF measurement HERE:

Denon PerL Group Delay with Personalization On

Denon PerL FRF Magnitude & Phase Spectrums (Visual representation of transfer function)

If you are interested in further discussion on transfer functions and how we can emulate a device’s sound personalization DSP, chime in.


Questions for consideration:

• What’s your conclusion on the devices’ performance, both the hardware and the hardware + DSP?
• Can we measure the DSP embedded in an earbud? If we can measure this, how should we use it to parse the hardware’s performance from the signal processing performance?
• If the hardware performs poorly, does DSP matter to you? Or it is hardware first?

 

[isostasy]

This is a great write-up thank you! Would it be at all revealing to complete the Masimo test in your own ears then measure on the GRAS to see how your own in-ear frequency response differs to the GRAS? You could then presumably use the difference to EQ your Blessing 2 (or any other IEM) to your own in-ear target frequency response?

 

[jjaskuna]

This is a great write-up thank you! Would it be at all revealing to complete the Masimo test in your own ears then measure on the GRAS to see how your own in-ear frequency response differs to the GRAS? You could then presumably use the difference to EQ your Blessing 2 (or any other IEM) to your own in-ear target frequency response?

Yes, very much so. The difference would provide insight into "your" personalized DSP vs the "ideal" dsp. This is assuming that the ear simulator is the "ideal" ear, which is a bit of a stretch. You'd need to do it in the frequency domain to get around the IR noise, but this would lend to calibrating any device's output to that of any sound personalization DSP. That is kind of the point I was hinting at: What if you could scale this form of DSP to the devices you want.