Over the years, I've spoken about using AI technology to help us understand the audiophile world (just a tiny piece among many, many other things to understand of course 😏). In early 2023 I discussed using LLMs like ChatGPT to generate content including the ease of making "subjective" text such as for product reviews. By 2024, we discussed the multi-modality of AI technologies; the ability to "imagine" not just words but also across generated visual and audio content. But then there are also silly uses like trying to sell snake oil.
While I think we have reason to be concerned about whether the financial model is going to work out with the massive AI infrastructure buildout currently under way, and the survival of many AI-related companies, IMO the technology and gains we've made with machines capable of thought is real even if we do have to ride out another Y2K-like Dot-Com Internet Bubble. It'll be painful (always is) for investors, but corrections are necessary.
Over the last year, we are now clearly in the midst of an "agentic AI" expansion where the public can now access virtual assistants to help us out with tasks beyond just chatbot Q&A interactions (OpenClaw being one that made headlines in November 2025, at that time called Clawdbot). Agents, like secretaries and assistants can do work for us behind the scenes like research, creation of reports, even do stuff like sending off E-mails or finding good deals on stuff to buy! We know that already more than half of Internet traffic these days is non-human agentic bots.
It's a brave new world that requires TRUST of course. For now at least, even though I'm OK with my AI agents seeing my work schedule and some personal E-mails, I certainly would not want to be the guinea pig giving it access to autonomously send off serious work without review or my credit card and bank numbers. 🤔
So, what does this have to do with audiophilia? Well, already it has been nice to get my agents to gather research data for blog articles. For example, a few months ago in April, I used my agent to gather data on inflation and examine the price of Stereophile Recommended Components. In September, an agent crawled across the Steve Hoffman Forums exploring what audiophiles are talking about these days. And now as we approach the end of 2026, we can access even more full-featured agents at our disposal that are capable of creating full briefings and analyses in ways that would take hours if not spread out over days to complete.
As I've noted over the years, I like Stereophile and the measurement data they've published across the decades on "High-End" products. It gives us a look into this "class" of devices that most hobbyists would not be able to test out for ourselves. In particular, large, expensive, loudspeakers are not the kinds of things we would routinely bring in and out of our homes for auditioning/testing!
To me, this library of loudspeaker measurements is John Atkinson's most important legacy since it was him who measured the majority of these products. The published body of work will probably be studied for years to come. (Unlike purely subjective reviews and opinions, mostly already forgotten.)
For this post, let's try something a little different. Let's dive into the "big data" and see if with the help of an AI agent, we can crawl across the Stereophile website, look for all reviews of loudspeakers of interest, extract measurement data, and find trends across the decades and among the brands.
You can imagine how painful this would be if one were not to use intelligent machines!
 |
| All Stereophile-reviewed loudspeakers 1964-2026 currently available online! Impressive body of work. [8K image.] |
 |
| All Stereophile-reviewed non-dynamic driver speakers from 1966-2026. [4K image.] |
As discussed previously regarding the most important components that make an audible difference in a sound system, other than the room and the source media, the next most important piece is without any doubt the loudspeaker. After all, it is the loudspeaker that converts electrical signals to acoustic waves - a lot can go wrong in this energy transduction process!
Pointing the AI agent at Stereophile.com, it was able to find 835 loudspeaker reviews since the publication began. While that's a good number, obviously Stereophile cannot review every hi-fi product made, so we must be mindful of potential bias for which products are submitted and accepted for publication - possibly related to relationships with companies, which companies felt they wanted to position their product in Stereophile, reviewers' preference, etc.
Not all reviews had measurement results, or a full data set. We must also be mindful of the limitations of the equipment used to gather the measurements. Most of the loudspeaker acoustic measurements were done with the venerable DRA Labs' MLSSA that started life back in 1989 as a MS-DOS PC app using time-gated quasi-anechoic technique. Electrical impedance measurements more recently used the Dayton Audio DATS V2 while older impedance results seem to be gathered with the Audio Precision System One.
To create that initial broad overview, let's have the AI agent explore the data, and start reading through John Atkinson's comments (mainly JA, but there were others over the years) in the measurement portion of Stereophile's reviews, capturing the numbers he highlighted as important, and creating a summary report of the corpus:
--- Loudspeaker Summary Report (1977-2026) ---
Here are the highlights:
Stereophile measured significantly more larger floorstanding (~507) speakers than standmount/bookshelf (~327) models.
Most speakers were built with dynamic drivers. See the image above of all the ones with non-dynamic drivers like electrostatic, planar, ribbon, AMT, horn or bending-wave technologies.
Median sensitivity did not change significantly - maybe a tiny increase (+0.049dB/y). However if we looked at it across decades, it does seem that more recent loudspeakers in the 2020's appear more sensitive.
Median minimum impedance stayed about the same in those impedance graphs - around 3.5-4.0Ω - over the decades.
Floorstanding speakers on average had +2dB higher sensitivity compared to standmount/bookshelf speakers. This stands to reason with more drivers in larger enclosures; for example, designing woofers in parallel likely will increase output by doubling cone area but likely lower impedance.
Qualitatively, ~80% of the loudspeakers measured were described (ie. by John Atkinson) as having "even" and "controlled" directivity as opposed to narrow or uneven.I don't think there's anything particularly surprising in the above and it's always good to have data; for example, the average higher sensitivity of floorstanding speakers.
While descriptions of the data are good as a summary, I think for readers who like looking at results, we know that what's most interesting isn't usually to be found in the text but in the detailed graphs that are posted. On a number of occasions, I thought Atkinson had a tendency to "give a pass" to products with poor results in his descriptions; so it's good to read between the lines and even better to interpret the graphs ourselves.
The B&W speakers can also be difficult for amplifiers as many of them also have excursions below 2Ω such as a number of the flagship 80X series especially around 50-100Hz.
Here are the headline findings and extra details:.jpg)
Since Stereophile publishes reasonably detailed graphs, let's ask the AI agent to go deeper - for example, let's analyze the impedance graphs by having the machine retrace them (in essence ingest/understand them), and then deriving the EPDR (equivalent peak dissipation resistance) curve for us, which combines impedance and phase to better represent load on the amplifier.
The calculation for EPDR used in Stereophile's measurements is unclear although reference has been made to Eric Benjamin's 1994 paper, "Audio Power Amplifiers for Loudspeaker Loads" (JAES Vol. 42 No. 9). For speed, the analysis in this post used an approximation: EPDR = |Z| / (1 + 4.2·|φ|/90°) which can be conservative and underestimate the EPDR by ~10%. ie. the actual loudspeaker load might no be as bad.
The official formula as per Jack Oclee-Brown is:EPDR = |Z| [(1-sin(5π/6))sin(5π/6)]/[(1-sin[(5π/6)+(2∣φ∣/3))sin(5π/6 - φ/3)]
As an example, we know some Wilson Audio loudspeakers can be hard to drive - here's a comparison graph of the Wilson Audio Alexx (2017) and the Alexx V (2021) with derived EPDR curves:
Impressive work by the AI not just in creating the retrace, but also taking the steps to confirm that what it calculated was consistent with JA's written comments. Let's ask the bot now to look at every Wilson Audio speaker with published impedance graphs and generate the EPDR where there is both impedance and phase data:
 |
| 4K image, click to enlarge/zoom. |
Notice in particular how punishing of amplifiers the Alexx and Chronosonic XVX models are with EPDRs substantially below 2Ω through much of the audible frequency range! For models such as these, it's obviously recommended that owners use only high current, low output impedance amps for best performance.
For comparison, here are all the Bowers & Wilkins (B&W) speakers Stereophile measured since 1987:
The format of the Stereophile measurements column has changed over the years. By 2012, the online publication increased the size of their graphs to 600-pixels which improves analytic precision. With AI as a research tool, let's direct it to examine the 2012-2025 Stereophile loudspeaker dataset by tracing the graphs in detail, and exploring whether there have been significant changes in the last 14 years.
Since Stereophile focuses on "audiophile"-level gear, we have a unique opportunity to examine how these higher-end products perform, including "price-no-object" models. The AI found that within this sample of loudspeakers from 2012-2025, on-axis quasi-anechoic frequency response measurements yielded 'flatness' between 200Hz-10kHz of +/-2.98dB for floorstanders, and +/-2.92dB for standmounts - slightly flatter smaller standmount speakers.
Across the 96 floorstanders and 89 standmount dynamic loudspeakers measured, here's what their average frequency response using a density color map looks like:
.jpg)
Over those 14 years, the AI was not able to discern meaningful progress in flatness of the measured frequency response across time, it found no discernable change in spectral tilt nor shift in speaker load. Notice in the frequency response density map above that most loudspeakers perform quite flat in the midrange from 300Hz to 3kHz. Below and above that midrange are where speakers deviate more.
While 'flatness' has been stable, bass extension has moved over the 14 years among the Stereophile-measured loudspeakers: the -6dB point for standmounts dropped from 43.9 to 33.8 Hz (n=30/21, p=0.001) and floorstanders from 31.0 to 25.7 Hz (n=30/28, p=0.08) when comparing 2012-2014 vs. 2023-2025 reviews. This is interesting.On the whole, as expected, floorstanders do provide more bass response, the -6dB bass point is about 11Hz lower than standmounts.As you can see with the EPDR graphs of the Wilsons and B&W's, actual speaker loads are harder than the minimal impedance number might suggest. Larger floorstanding speakers tend to be more difficult loads - median minimal impedance of floorstanders = 3.27Ω, and standmounts = 4.23Ω; a whole 1Ω reduction for the floorstanding speakers.
The AI analysis made an interesting comment about the overall tilt and frequency response errors: Tilt is level on average, and response errors cluster in two bands. Median slope changed an insignificant +0.09 dB/decade: 150 of 275 within ±1, 67 rise more than 1, 58 fall more than 1. 158 speakers have a deviation over 3 dB from their fitted line. Most common response error clusters are treble peaks at 8–16 kHz (90 of 290 events) and dips at 2–6 kHz (70).
Given what looks like little change other than speakers measured from 2023-2025 having greater bass extension than those from 2012-2014, I wondered if there's any evidence of change if we look even further back.
Let's start with loudspeaker impedance and EPDR and comparing loudspeakers from 1990-1995 vs. 2020-2025, we can indeed see that loudspeaker loads have dropped on average over the decades:
In particular, notice the top right cumulative curve showing that newer loudspeakers from 2020-2025 (red) tend to have lower EPDR minima than those from 1990-1995 (blue). Numerically, the median EPDR minimum fell from 1.95Ω in 1990–1995 → 1.72Ω in 2020–2025 (p = 0.007, r = −0.22).
Now, let's compare the average frequency response of loudspeakers from the decades 1990-2000 vs. 2015-2025:
Interesting. Visually, we can see that the 2015-2025 models have on average a slightly smoother midrange and low-treble (overall flatness 200Hz-10kHz dropped from ±2.9 → ±2.8dB).
However, the older floorstanding speakers from the 1990's on average had slightly deeper bass extension than models in the last 10 years. Models from the 1990's had a median -6dB point at 25Hz compared to 30Hz for 2015-2025 speakers in this dataset (p = 0.001). What this suggests is that back in the 1990's, Stereophile probably measured big, deep-bass, designs that probably slipped in the early 2000's, but more recently from 2023-2025, the magazine has again measured speakers with deeper bass capability compared to 2012-2014.
A different bass extension picture emerges when we compare standmount speakers from the 1990's with models from the last decade:
Modern standmount speakers on average can dig a bit deeper into the bass compared to those of their earlier counterparts with -3dB point going from 49 → 44 Hz and about +1dB "fuller" sound from 40-100Hz. Notice that modern standmount speakers also seem to have flatter "presence" 3-8kHz region.
In all these frequency response measurements regardless of decade, we see a ~10kHz bump. This is most likely an artifact of the Stereophile measurement system. As a result, accuracy of the loudspeaker data 8-11kHz is suspect.
Unfortunately, the Stereophile data does not contain information about distortion.
However, we can look at other characteristics like directivity - here's a mosaic of the horizonal off-axis directivity for every loudspeaker with data from 2012-2025:
 |
| Lateral dispersion from 500Hz (left) - 20kHz (right). Normalized 0dB on-axis (bottom), and 90° off-axis (top). Large 8K image, click on image and zoom in for a better look. |
While many of the thumbnails are complete, the grey zones are areas that the AI was unable to trace out based on the published graph. At times the data is absent or represent "dense" regions of variability where the AI did not have adequate resolution to be confident when trying to convert to a flat representation.
We can imagine these patterns as unique "fingerprints" of how the loudspeaker would interact with the room. For illustration, let me pull out 3 interesting examples:
Whether one is an objectivist or subjectivist, I think all of us would agree that if we were to put these three speakers in the same room, at the same location, the sound would be quite different! Not only are they based on different driver technologies (MBL's radialstrahler, Klipsch horn, Totem dynamic drivers) but we can imagine the off-axis radiation patterns interacting with the room in significant ways - the MBL's omnidirectional pattern would be different from the Klipsch's narrowing at high frequencies and the Totem's unusual on-axis hole around 2kHz making the off-axis look abnormally accentuated, telling us that for these speakers, it would be essential for the listener to sit off-axis for best results.
Although not typically as important as horizontal off-axis data since out ears are arranged laterally and side wall reflections tend to be stronger, vertical off-axis data is still worth examining as it will affect ceiling and floor bounce as well as determine the effect of seating height above or below the tweeter. Here are the same speakers' vertical directivity data reconstructed:
Again, we can imagine the unique speaker-room interaction playing out as a function of these lateral and vertical directivity patterns across 3D space and time not just from the speaker's radiation pattern but also the room's modes and reflections moderated by absorption and diffusion, plus where the listener is seated.
Like with the frequency response and impedance discussions above, I also wondered whether there is evidence of improved off-axis performance if we compared devices from years ago with contemporary products. Here's an analysis of lateral off-axis response from products reviewed between 1990-1995 and 2020-2025, more than 50 products in each group:
It doesn't look like there's a big difference despite the 30-years span. However if we dig into the statistics, newer loudspeakers did have a smoother era-median graph and the 30° off-axis smoothness was calculated to be statistically better. Only with listening tests can we know whether this statistically significant change at a group level actually translates to it being audible.
Audiophiles often will talk about time-domain performance. Sometimes, I think this is taken to the extreme as if the ears are highly sensitive to very small changes. It's interesting though that while step-responses can look quite different, I don't believe there's good evidence that the sonic correlation is anywhere as strong as frequency response. For illustration, here are a few side-by-side step response graphs of loudspeakers over the decades - floorstanders and bookshelves:
Yes, quite different, but making audible correlations by looking at these step responses plotted on millisecond time scale would be difficult to say the least! Having said this, we can still apply time-domain DSP optimizations as part of our sound room set-up at the listening sweet spot.
[JA has written about the lack of correlation between time-coherence and audibility: "Of the 350 or so loudspeakers I have measured, there is no correlation between whether or not they are time-coherent and whether or not they are recommended by a Stereophile reviewer."]
For completeness, here are the most recent 16 time-coherent loudspeakers (a single positive step with no staggered driver arrivals) from 2012-2025. Seek out these loudspeakers if you want to have a listen yourself and whether you think time-coherence makes a difference for you. As you can see, time coherence is achieved using different methods:
First-order crossover designs. Vandersteen is the only brand that shows up repeatedly, which makes sense given Richard Vandersteen's long commitment to time and phase coherence.- 2013 Vandersteen Treo- 2016 Vandersteen Model Seven Mk.II- 2019 Vandersteen Quatro Wood CTFull-range drivers with no crossover in the critical band- 2013 Audience ClairAudient The One- 2016 Zu Audio Soul Supreme- 2019 Manger p1 (bending wave)- 2020 Rethm Maarga- 2022 EJ Jordan Marlow- 2024 Voxativ Ampeggio 2024Electrostatics- 2012 Quad Reference ESL-2805- 2017 Quad ESL-2912DSP-corrected active speakers. Their time alignment is done in software.- 2016 Dynaudio Focus 200 XD- 2017 Kii Audio Three- 2019 KEF LSX- 2025 System Audio Legend 40.2 SilverbackThere's also one coaxial: the 2022 Tannoy Stirling Prestige Gold Reference, a Dual Concentric with only 0.09ms of stagger.The other 94% of the 277 speakers show the usual staggered arrivals: tweeter first, then woofer, often with a polarity flip.
As for the cumulative spectral-decay (CSD waterfall) plots (a hybrid time and frequency domain measurement), it was more difficult to analyze the Stereophile data on account of variability of range (floor) between 15-33dB and the time window used by JA over the years. Furthermore, many of the reviews published before 2000 were of lower resolution. In theory, over time, we could expect improvements from better cabinet designs and bracing using better computer simulations. The best apples-to-apples comparison I could make was to match floorstanding (larger, more potential stored energy) dynamic loudspeakers across 20 years from 2000-2005 vs. 2020-2025; here's the analysis and below we have representative waterfall plots of speakers with faster to slower decay from each timeframe:
Overall, comparing the loudspeakers across 20 years between 2000-2005 (54 products) and 2020-2025 (66 products), there was no evidence of significant group difference on the waterfall plots. All we can say is that some speakers have good decay times, others don't, but on average, we're not seeing evidence that technological advancement has meaningfully improved the decay times from 1kHz up. Who knows, maybe by 2000 the speaker cabinets were already excellent in the audiophile market and we would have needed to look at older products to show the improvement. Alternatively, perhaps improvements were below 1kHz which we can't accurately test due to the gated time window measurement technique.
In Closing: Progress over time? What were the best-measuring loudspeakers?
Looking through this data, I was left wondering - "Have there been clear improvements in loudspeaker performance over the decades?"
I believe the answer is "yes", there have been improvements, but I think we must also appreciate that much of loudspeaker technology has already matured. I see no evidence of major change in performance in the last 30 years; improvements have been evolutionary, not revolutionary. Changes have been in the form of small shifts in sensitivity and variation in bass extension like those described above. In theory, we should also see improvements in smoother off-axis performance over the last couple of decades as "spinorama" measurements (CTA-2034) have become standard evaluations resulting in better crossover and waveguide designs (we probably saw some of this in the 1990-1995 vs. 2020-2025 lateral off-axis comparison above).
Since Stereophile caters to a more high-end audiophile crowd who prefers component systems, allowing them to mix-and-match ostensibly higher quality pieces, this data mostly reflects passive loudspeaker performance. Among the 288 speakers analyzed in detail from 2012-2025, only 22 were partially if not fully active designs. Innovative active loudspeakers incorporating DSP tuning could bring another level of sonic control for the end-user and potentially better sound can be achieved through driver-crossover-amp synergy. Perhaps we'll see more active products reviewed and measured ahead?
While it's great having access to all this data from across the decades, it's important to appreciate that the Stereophile measurements are limited. This data set is based on quasi-anechoic gated measurements at 50" which limits resolution below ~250Hz and the spliced nearfield driver + port response introduces potential level-matching errors and may be ignoring phase cancellations. That single 50" measurement distance also could be suboptimal for large speaker and line-array integration. Would have been great if all these loudspeakers were measured with something like the modern Klippel Near Field Scanner of course, but I think we've already learned a lot already.
A major limitation is the absence of distortion data and we don't know how loud these loudspeakers can play without breaking up. While we cannot test the hypothesis using the Stereophile data, it would be reasonable to think that over time, modern loudspeakers can achieve lower distortion with better motor designs. Diaphragm material improvements might also play a role in this, pushing break-up modes further outside of intended operating frequency range.
Practically, let's also not forget about improvements in manufacturing compared to years past. Mass production improves consistency, and optimizations in production efficiency also result in potential cost savings and access to higher-performance loudspeakers over time - the usual trajectory for high-tech products (as opposed to luxury brands and the Veblen economy). Stereophile's reviews are typically unable to shed light on the high-value consumer segment.
There's good news I suppose in NOT finding massive differences over the years in these objective results. It means that 'vintage' pre-2000 loudspeakers, can still sound competitive compared to modern products assuming of course they're in good working condition. If you're buying used, make sure to inspect and fix any rotting foam woofer surrounds, replace off-spec electrolytic capacitors in the crossover, update tweeter ferrofluid, clean oxidized connectors, etc.
Since the AI agent has ingested all this data, let's end off with an AI-assisted assessment of the best loudspeakers measured by Stereophile from 2012-2025 using the following criteria:
- Mandatory: EPDR >2Ω - good speakers must be easier to drive, including tube amps.
- 5-way equal-weighted average of:
1. Flat on-axis quasi-anechoic frequency response (average RMS deviation).
2. Smooth on-axis range (minimal peaks, dips, and treble excesses).
3. Smooth off-axis performance (30°) - more predictable speaker-room interaction.
4. Lower bass extension - definitely better for rock & modern electronic genres!
5. Higher sensitivity (dB/2.8V/m), expand amplifier options (eg. low-power amps).
Best Floorstanding Loudspeakers
#1. JBL 4367 Studio Monitor: (US$16.5k/pair) On-axis flatness from 200-10kHz is 0.79dB, top 22% of all loudspeakers tested with 30° off-axis smoothness 1.5–5 kHz top 5%ile. Bass -6 dB point is great at 29.4Hz and its EPDR stays above 2.4Ω. Reasonably high 92.7dB measured sensitivity.
#2. Spendor A7: (US$6.3k/pair) On-axis flatness from 200-10kHz is 0.77dB, top 19% of all loudspeakers tested with 30° off-axis smoothness 1.5–5 kHz top 12%. Bass -6 dB point at 32.8Hz with EPDR 2.75Ω. Measured 87.1dB sensitivity is unfortunately a bit lower compared to other floorstanders.
#3. EgglestonWorks Viginti: (US$39k/pair) Tight on-axis range at top 12%ile, much deeper bass with -6 dB at 24.3 Hz than the Spendor A7, even though its 30° curve is rougher. Decent 89dB sensitivity.
Honorable Mentions: Ignoring the EPDR, the KEF Blade Two Meta (US$28k/pair) tops the list with excellent 3%ile on-axis flatness, 11%ile off-axis smoothness, -6dB bass at ~28Hz but low sensitivity of 86.7dB measured. It's a tough load though with minimum EPDR down at 1.23Ω at 34Hz. The Estelon X Diamond Mk II (US$89k/pair) enters with the deepest bass in the group (-6dB point at 19.7 Hz), but it dips to minimum EPDR of 1.8Ω. On-axis flatness is only 34%ile but top 8%ile for 30° off-axis smoothness.
Best Standmount / Bookshelf Loudspeakers
#1. Harbeth Monitor 30.2 40th Anniversary: (US$6.5k/pair) Very flat frequency response with on-axis flatness from 200-10kHz of 0.56dB - top 4%ile - and 30° off-axis smoothness from 1.5–5 kHz top 10%. Minimum EPDR calculated at 2.8Ω at 563 Hz. Its bass -6 dB point of 40.6 Hz is a limitation but not too bad and can be supplemented with a good sub. 87.3dB sensitivity is in the top third of standmount speakers measured.
#2. Tekton Design Impact Monitor: (US$1.7k/pair) Good on-axis flatness at 9%ile and 30° off-axis smoothness from 1.5–5 kHz top 22%. Bass -6 dB at 34.5 Hz is excellent but minimum EPDR 2.01Ω at 92Hz so just passes the mandatory criterion. 87.5dB measured sensitivity puts it at 27%ile.
There has been some bad press about this company a few years ago, as we also discussed here. Technically though, the measurements look good based on JA's work although I know there have been criticisms about possible intermodulation and phase interactions with the multiple small drivers acting as midrange array around the center tweeter.
#3. Mission 770: (US$5k/pair) Not the flattest on-axis at 44%ile but 30° off-axis smoothness is good at 15%ile. Remarkable bass for a standmount (-6dB point at 27.3 Hz - 5%ile, among the deepest of standmounts) and minimal EPDR at 2.99Ω. Measured 85.5dB sensitivity is a little low so you will need a reasonably powerful amp for house parties.
Honorable Mentions: KEF LS50 Meta (US$1.6k) looks good overall with excellent 9%ile flatness and 6%ile off-axis smoothness, but sensitivity 84.5dB lowish (78%ile) and bass with -6dB point at ~42Hz puts it down at 68%ile for standmounts. The related KEF Q350 (US$500) performed well also - a kind of "budget LS50" but EPDR drops down to 1.65Ω and there's more enclosure vibration.
I think that looks like a reasonable list of loudspeakers to consider if objective performance is a factor in your pursuit for good sound and you want the speaker to be able to pair nicely with most amplifiers. Notice that using a complex balanced multi-way criteria like this, clearly there is no such thing as perfect loudspeakers and we need to decide for ourselves which characteristic(s) we value most when making a purchasing decision.
One final list: suppose we have a top-of-the-line amplifier with high power and strong damping capability, and we don't need to care about sensitivity, don't care if floorstander or standmount, the price, or difficulty to drive (ignore impedance & EPDR). Which are the top 10 highest-ranking loudspeakers based on Stereophile's measured products from 2012-2025 using only frequency-domain criteria: flatness and lack of local bumps and peaks on-axis, deep bass performance, plus smoothness off-axis?
#1. KEF Blade Two Meta (US$28k/pair, 2022)#2. Revel Performa F228Be (US$11k/pair, 2019)#3. KEF Reference 5 (US$25k/pair, 2017)#4 (tie). JBL 4367 Studio Monitor (US$16.5k/pair, 2022)#4 (tie). KEF Blade Two (current pre-owned ~US$11.5k/pair, 2015)#6. Estelon X Diamond Mk II (US$89k/pair, 2025)#7. Vivid Kaya 45 (US$18k/pair, 2020)#8. Thiel Audio TT1 (US$6-7k/pair, 2016)#9. Rockport Avior II (US$38.5k/pair, 2017)#10. GoldenEar Triton Reference (US$8.5k/pair, 2018)
Keep an eye/ear out for these brands and models the next time you're at an audio show as examples of technically well-engineered products. Notice that the Revel and JBL brands are under the Harman umbrella, and there are 3 KEFs on the list. We also see a wide variation in asking price but most of these top-10 products are listed well below US$50k; should be no surprise that price and sound quality need not correlate.
You might be surprised though to see the absence of "subjectivist audiophile favorite" brands that show up in recommended component lists and "best of" awards like Wilson Audio. Even though JA measured many of these (full data for Alexia Series 2, Sasha V, Sasha DAW, Sabrina, SabrinaX, Yvette, Duette Series 2 with MSRP ranging from US$16-58k/pair), none made it on the top-10 list. It's because in the measured data, Wilsons tend to have non-flat on-axis response that's slightly more "forward" due to depressed upper-mid and a low-bass hump with uneven off-axis response as the drivers often sit in separate, potentially adjustable, modules.
These kinds of anomalies will show up in measurements even though psychoacoustically, listeners might subjectively enjoy this kind of subtle EQ and Wilson does a great job in their advertising to high-end audiophiles. While Wilson advertises about having "precise time alignment behavior", these speakers with their passive crossovers are not time-coherent when we examine the step responses.
With access to the frequency response of 19 Wilson Audio speakers measured over the years, we can visualize the brand's median "house sound":
 |
| The Wilson Audio "house sound". Graph created using JA's "listening window" - average across ±15° horizontal and ±10° vertical, at 50" with nearfield woofer+port(s) spliced in below 300Hz. SPL not explicitly stated in reviews, assume ~80-90dB (2.83V). Each curve normalized 300Hz-3kHz and then overlaid. |
We can similarly analyze the beautiful-looking Sonus Faber speakers (JA fully measured the Aida, Sonetto V G2, Olympica Nova 1, Guarneri Tradition, and Lumina III models with prices ranging from US$2k-130k/pair). This brand typically shows uneven frequency responses with dips in the upper-mid region (~3kHz) and a tendency for "brighter" upper frequencies.
 |
| The Sonus Faber "house sound". Notice overall a smoother midrange with fuller bass and also accentuated upper end compared to the Wilson sound. A bit more "smiley face" EQ-like. |
Also notice with the Sonus Faber products the lower-mid boost from 300Hz to 1kHz. This is a little like the "Vocal Bloom Curve" I discussed a few months back which might add some sweetness or weight to voices.
 |
| As with Wilson, Sonus Faber step responses also are not time-coherent. |
Finally, let's look at a brand that has multiple products in the top-10 rankings above - KEF - using the pooled data of 12 of their products from 1996 to 2022, here's their median frequency-response signature:
Notice how much flatter the frequency response is for KEF compared to Wilson Audio and Sonus Faber. I believe this is representative of this brand's reputation for excellent engineering.
 |
| For completeness: 3 KEF step responses. |
So, let's overlay the Wilson / Sonus Faber / KEF / General Pool frequency response graphs:
I think this is a nice example of brand-specific insights we can "mine" from the Stereophile data to show that some loudspeaker makers actually do have a certain kind of "voicing" inherent in their products. Whether the sonic signature is truly something they were aiming for, or the outcome of the drivers they chose and cabinet shapes they tend to use can only be answered if we know the company's development process.
One last thing: We can create EQs that can "emulate" the brand's house sound. For example, if you have flat floorstanding loudspeakers that more or less follow the pool median frequency response, using a 5-item parametric EQ, you can make it sound more like the "Wilson Audio House Sound" with these settings:
Turn preamp down -2.2dB to prevent clipping.
# Type Fc Gain Q1 Peak 147 Hz −3.7dB 0.432 Peak 698 Hz −1.9 dB 2.673 Peak 1230 Hz −2.4 dB 2.544 Peak 2530 Hz −3.4 dB 3.105 Low Shelf 2900 Hz +2.3 dB 1.00
Or try this set of EQs for the "Sonus Faber House Sound":
Turn preamp down -1.2dB to prevent clipping.# Type Fc Gain Q1 Peak 277 Hz −2.0 dB 0.922 Low shelf 765 Hz +1.1 dB 1.003 Peak 994 Hz +1.3 dB 6.004 Peak 2810 Hz −0.9 dB 2.955 High shelf 9290 Hz +1.0 dB 1.00
Since the frequency irregularities are not as strong with SF as the Wilson sound, the gain/loss values are lower.
Visualization of these EQ settings:
Alas, the KEF sound is kind of boring because it's just neutral. No point making an EQ for this brand if you already have neutral speakers. 😉
I encourage you to play with the data and experiment, such as by plugging in the parameters into Roon's Parametric EQ DSP among others:
Happy November 2026, dear audiophiles. I hope you're enjoying the music!
Important week in politics ahead, as per discussions last week.
No comments:
Post a Comment