Warren TenBrook's Summary of Head Measurements at Harman
(Editor's Note: My head has been buried in the data for some weeks now. What a pleasure to come up for a little air reading Warren TenBrook's summary overview of his experience and thoughts about the process so far. A clear, concise, and enjoyable read. Thank you, Warren!)
I arrived at Harman International Northridge at the appointed time of 9 am, sharp. After being conducted to the Harman Reference Room, I found Tyll Hertsens already busily setting up his gear, focused on collecting all the necessary interconnects to begin his Head Acoustics simulator measurements. Tyll's next stop was T.H.E. Show, Newport later that day and he didn't want to waste any precious time.
I engaged in some conversation with Tyll, Todd Welti, and Omid Khonsaripour of Harman Acousting Research, and Henry Goldansky, JBL Professional Engineering. Harman showed keen interest in Tyll's project, and several employees assisted. I blended in, maybe too well. Tyll thought I was from Harman until we were properly introduced later that morning. Good for a laugh.
The purpose of my visit was to meet Tyll and the Harman staff, and better understand how the sound signature from reference loudspeakers translates through room acoustics to head simulator measurements and finally to headphones. Can a headphone sound like good speakers in a good room, and can we back that up with data?
My previous article summarized the Harman Reference Room acoustics, and Sean Olive's blog goes into some depth. The 500 square foot room is similar to a high-end home theater, mastering studio, or audio salon demo room; however, the speaker placement and acoustics are carefully controlled with RT60 0.4 to 0.5 seconds across the band with smooth bass throughout the room.
Harman has conducted head simulator measurements in the room with up to seven JBL Professional LSR6332 stand mount monitors in an ITU circle arrangement, with four subwoofers placed in the corners. This is the original standard loudspeaker compliment for the room. A stereo pair of Revel F208 floorstanding speakers were also measured in later studies. Tyll's measurements added the JBL Professional M2 monitors to the mix. Each of these systems has been designed for anechoic flat on-axis response and controlled off-axis and power response.
In the last post I emphasized that anechoic flat speakers exhibit a response slope when placed in the listening room, and there will be some response irregularities even with optimum acoustics. Harman has done considerable research on room curves to smooth the response in the listening area. The Harman Audio Test System (HATS) is used to apply response curves to speakers in the Reference Room.
Figure 1 presents several relevant speaker response curves. All my graphs are about 1/3 octave, 0.5 dB resolution, normalized at 200 hz. Unlike the smoothing functions used by Arnaud on Tyll's data, I let Excel smoothing do all the work. The results are for illustration only.
The curves fall into three families:
- The standard Reference Room (RR) curve (red) used in the 2013 Harman headphone study was developed using double-blind preference tests by trained listeners. A very similar curve for the M2 (green) was loaded in the speaker DSP for Tyll's head simulator tests.
- An alternative curve RR1 (violet) was also measured in the 2013 Harman study, described as "...generally preferred based on a previous study where different room correction systems with different target response curves were evaluated by trained listeners." Curve RR1 is similar to a 1 dB/octave slope (orange). Finally, the B&K 1974 empirical curve (grey) has less bass, but follows curve RR1 and 1 dB/octave closely above 200 hz.
Perhaps the classic B&K curve lacks the same elevated room gain in the bass as few loudspeaker systems in 1974 could be trusted to have flat extension to 20 hz. The curve used for Tyll's measurements and the standard RR curve each have progressively more bass and treble. Whether this is more "accurate" boils down to weight we assign to the blind testing preferences underlying curve RR.
Sean Olive was kind enough to link to a comprehensive paper by Dr. Toole that provides another view of in-room speaker curves. In figure 13(b) the Revel F208 and JBL M2 room responses are presented with a predicted "home and professional monitor loudspeaker with flat direct sound responses - estimated steady-state sound in a typically reflective room" (heavy broken line):
The trends suggest that speakers like the Revel F208 and M2 will tend to naturally follow curve RR1 and 1 dB/octave slope within the limits of their bass extension when placed in a good room, discounting the bass irregularities Toole attributes to standing waves.
I also note that Figure 4 in Olive-Welti-McMullin AES paper 8867 (2013) presented headphone listening tests where corresponding curve RR1_G was preferred over curve RR_G. Presentation slides summarizing the preferences were posted by Dr. Olive.
In my opinion, the evidence weighs in favor of curve RR1 as a neutral reference curve over the typical Reference Room house curve, RR, used during Tyll's measurements.
How would these loudspeaker curves measure with a head simulator? Toole, et. al. developed techniques to translate anechoic measurements to predicted loudspeaker response at the listening area in real rooms. In the same spirit, Figure 2 illustrates a predicted headphone response for each of the speaker room curves in Figure 1. The predictions are the simple sum of the Figure 1 speaker curves and Harman's GRAS 43AG head measurements using a stereo pair of Revel F208 speakers adjusted flat in the listening area. The Figure 2 predictions help fill gaps where there are no head simulator measurements for a given speaker curve.
- As expected, curve RR (red) and the M2 curve used for Tyll's tests (green) are similar, and are predicted to have the most bass and treble emphasis.
- Curve RR1 (violet) and 1 dB/octave slope (grey) again follow each other closely, with a bit of bass rise often seen in closed-back headphone designs (think Oppo PM-3 or Ether C).
- The B&K 1974 curve (grey) generally resembles high-end open back planar headphones with extended bass (think HiFiMan and Audeze).
- All the curves show a gradual slope in the midrange (again PM-3, Ether C, or perhaps HD650)
Figure 3 presents head simulator measurements. Tyll's Head Acoustics independent-of-direction curve is also included.
The curves follow the general trends of the predictions, but details vary:
- Tyll's Head Acoustics 2016 measurements (all directions summed - green) and Olive-Welti 2013 RR curve (red) are similar, being based on almost identical speaker calibration. The Head Acoustics midrange response is lower than the GRAS 43AG followed by more sharply-defined 3 kHz peak.
- Modified curve RR1 (violet) is more typical of the publicized 2013 "Harman Curve." It has less bass and treble emphasis than Tyll's measurements, as predicted.
- Tyll's Head Acoustics simulator independent of direction compensation curve (gold) lacks the bass emphasis of curves RR or RR1, but the general shape of the treble response is similar to Tyll's measurements of the M2. The pattern speaks to the personality of the Head Acoustics design.
More number-crunching on Tyll's data may alter the picture slightly, but Tyll's final Head Acoustics results are likely to retain a family resemblance to the RR-type curve used for the JBL M2. The type of curve applied to the speaker system may well explain Tyll's reservations about the measured bass response. The 1 - 2 kHz inflection seems to be more characteristic of Head Acoustics design.
I think we can state the 43AG and Head Acoustic systems respond a bit differently to the common RR loudspeaker curve, but there are some clear 'Harman Curve' tendencies in Figure 3:
- Bass should be as well-extended as possible, and evidence supporting some rise in bass level below 200 hz.
- Gentle midrange increase above 200 hz.
- 3 kHz peak 10 - 13 dB above baseline.
- Return to baseline level at 10 kHz.
So have we arrived at convergence or consensus? If you stripped the labels from Figure 3 and asked readers if they were the same, or even similar headphones, I think the results would be mixed. There is more work to do. The B&K 1974 study alone investigated five loudspeaker systems in three rooms. Measuring with the Head Acoustics, 43AG, or other simulators in more treated rooms would be a plus. For example, measuring the 43AG in the Harman Multichannel Listening Lab and the Head Acoustics in Bob Katz' Orlando mastering studio via Tyll's association might be instructive.
Like many topics in audio, headphone curves are controversial. Tyll and I seem to agree head simulator measurements could lead to a good compensation curve. But there are dissenters who assert headphone target curves are "...Over reliance on scientific models, AES papers, German overthinking and over-complexity…". During conversations with Harman, it is clear they also feel the tension between a by-the-numbers 'Reference' approach versus a listener 'Preference' approach. They reconcile the two philosophies by using proven scientific techniques to outline 'Reference' objectives, and measure 'Preference' for competing proposals using blind testing with trained listeners.
My wife and I are both in scientific fields, but are musicians as well. When I returned from Northridge, she skeptically asked if this work could get us any closer to a live concert experience?
I replied that the closest event to the listener in the original recording chain is the mastering room. The mastering room also bears most resemblance to a home listening room. I added that many recordings are complete artifacts of the studio and are never heard in a natural acoustic space like a traditional concert hall or club. They are only given life when played back during mixing and mastering. The mastering room is the only common acoustic target, so it makes sense to measure and attempt to replicate its sound over headphones.
I offered that binaural recording could get us closer to live sound, but very few binaural recordings are done - it's a niche with its own limitations.
Then it occurred to me that Harman and Tyll's head simulator measurements are capturing a tonal signature akin to a binaural recording in a good stereo mastering environment. Perhaps there will come a day when authentic binaural tracks for headphone listening could be imbedded in multichannel audio recordings for a 'You Are There' perspective. Until then, striving to attain a spectral balance close to a good mastering studio is the next best thing.
I want to thank Dr. Sean Olive and Tyll Hertsens for this opportunity. I appreciate Todd Welti, Omid Khonsaripour, and Harman staff generously offering their time and access to Harman's facilities, including the Multichannel Listening Lab and Eargle Theater. It all added up to an unforgettable experience.