If I can be permitted to provide a few observations for balance, some of you may want to read the following on the topic, all IME but with a smattering of objective as well as subjective caveats.
As with many such articles, it's mostly subjective narrative with no evidence backing up claims of microphonics at audio frequencies which whilst easy enough to measure has nothing in that article to reference the context nor magnitude. I did some work about 5 years ago on microphonics in poly caps at audio frequencies, and it will come as little surprise to those who have tested these things to learn that it was so low as to be utterly inaudible. less than a tiny fraction on 1dB...and utterly inconsequential in things like passive crossovers for example.
The "up" curve of the ESR also does NOT happen at audio frequencies but higher up at radio frequencies, so again for just about all audio applications except where UHF/VHF decoders or similar are concerned, all caps behave in quite a predictable fashion and effectively have an asymptotically lowering of ESR as frequency increases, so whilst esr impedance starts high, you can take a cap of any sort of the same value and they will all follow the same impedance profile to well above audio frequencies. I plotted a series of readings at 1KHz, 15KHz and 200KHz for a range of caps and whilst at 200KHz the values varied from about 0.5 Ohms to just 0.05 Ohms ESR, that variation was more to do with plate dielectric choice so the 'lytics had the higher reading.
In circuits that use a lot of negative feedback may wonder why more manufacturers don't use it in valve amps as the holy grail is seen as zero negative feedback. Where output transformers are used, for fidelity and control, this is often the worst solution for relatively high inertia cones producing back emf which means that power has to be dissipated within the amps output stage leading to poor damping and a tendency of the output at the speaker to follow the curve of it's impedance profile and become non linear and "woolly" sounding. This is because output transformers have a high output impedance due to the impedance of the windings on the secondaries. So how is that controlled? Using global negative feedback which takes part of the output and feeds it back so providing a lower output impedance which whilst not quite as efficient does gain more control but at the expense of amplifier stability.
It's this stability control which makes implementing such circuits properly challenging and often a little negative feedback is actually worse than none at all, so manufacturers use none or perhaps just 5dB which is nowhere near enough. Radford designs overcome this as Arthur Radford designed a feedback loop that didn't oscillate too badly at frequency extremes meaning that he could apply about 27 or 28dB of feedback with no loss of amplifier stability and vice like bass grip. In such feedback loops it is partially the electrical resonance of capacitors which add to this problem at frequency extremes both above and below that which might be audible but which extend the influence into the audible band creating increased distortion and amp instability. Radfords created a clever design to control these resonances at frequency extremes.
In crossovers, be they passive or active, there is no danger of these resonances occurring at the the frequencies concerned, none, so marketing of foo brands is just that....marketing. You can swap a high end paper in wax megabucks 3.3uf for your tweeter with a modest claritycap esa manufactured by IPL in Wales(a quality capacitor) and there will be zero difference in frequency response, slope or distortion...zero. Think about that before utterly wasting your hard earned.
Some larger rated caps behave slightly more differently in crossovers simply due to stabilisation times being longer, and in principle (although insignificantly audibly) measured distortion could be a bit lower but we're talking the difference between say 0.01THD to 0.05% THD. No one can claim that is audible or significant in any way. The settling times can be though as a 630v cap being fed a 12 to 15v signal may take much longer to stabilise hence people often hear more differences between the two initially but if given say 20 or 30 hours running, they'd likely not hear any difference. I've gone through many hundreds of pounds worth of caps of a specific value in my speakers seeking out which, if any, make that longer term difference and whilst there are some unexplained differences in a very few (yes it's audible) it just means we're measuring the wrong things. It is a rarity though.
There is one area that few know or talk about though which can be of greater significance to resonance, especially when you take a component which can at certain frequencies exhibit resonant behaviour and place it within a particularly electrically resonant part of a circuit. I discovered this almost by accident but there are a few engineers I discussed this with who chuckled and explained possible reasons as they were aware that it was a possibility. This has to do with plate design, how tightly the plates are wound and dielectric material.
The best performing (audibly) caps within this specific electrical circuit found in many crossovers all shared one trait...they were short plate, single plate tightly wound designs which just gave a cleaner sound over the band they were helping control and with a slightly lower distortion profile.
Working backwards I could sort of work out why but still could and cannot explain it in any mathematical or physics means...it just is what it is until I can! My theory on this has to do with plate distortion upon applied AC current since this was mostly observable in AC circuits. It stands to reason that when energising a cap one of the things that happens to a greater or lesser degree can be micro-distortion of the plate upon applied current. It is fractional but nevertheless happens and varies depending on physical properties of the cap including plate width and dielectric density plus winding stress and of course significantly, applied current. It stood to reason that that cap that performed best in the resonant circuit was a short plate highly stressed winding.
The converse was tested with a pretty ordinary budget twin plate design (two plates usually wound side by side to make up the required capacitance within a dielectric). A past client of mine called me to his home where he had a mystery that he couldn't solve....with his amplifier switched off so no direct route for powering the speakers, he switched his valve (Audionote) DAC on and passed a signal through it from his streamer. We both clearly heard music! Opening the DAC up, we discovered the source...it was large Mudorf Z caps acting as coupling caps and they were merrily singing away to the music due to the effects of applied plate distortion under electrical alternating load. I've never heard the same since from any cap. I tried the reverse, measuring outputs to see if this cap behaved like a condenser microphone, but it just didn't which sort of rubbished what KH suggested in that article. It would take so much vibrational energy to even remotely create a measurable, let alone audible cap distortion that the amount of energy needed would likely cause your home structural damage before a reading was obtained and your eardrums would have burst along with it!
So whilst this cap could not be said to be microphonic at all, the very fact that under applied current it sang, does mean that part of that signal is lost in the power that was wasted creating plate distortion and audible resonance so the only conclusion here is that it was the wrong cap for the job, and at the prototyping stage, alternatives ought really to have been tried to find one which didn't exhibit this behaviour. If plates were distorting that much under the high voltage current these were receiving then it stands to reason that distortion must have been affected too here as well as circuit efficiency.
What I learned from all of this was a number of things which are taught in 101 electronic design: Not all caps are made the same, some behave differently than others at VHF and UHF frequencies, plate design can be relevant in electrically resonant circuits and rating can affect stabilisation times. ESR differs a little but not as much as most people would like to believe. Box polys commonly found in amps and other audio circuits have very low esr and quality bipolar electrolytics also have very low esr (I measured one a few weeks ago at 0.15Ohms ESR at 200KHz).
Please take what you read on these articles with a big pinch of salt because context is everything and that chap has not properly added context, I suspect deliberately for an article that was meant to provoke thought. Rather than believe it and rush out to buy the latest super duper cap, take advice of people like Alan, Will (from Radford) or similar as they really are the people best placed to offer that advice.
I am happy to provide advice on cap choice for crossover work and this is borne both of experience and of testing probably way more than anyone on here has handled or is likely to, because I didn't take my approach to design lightly and have invested heavily to approach these things with an open mind but also with basic electronic knowledge of the behaviour and design of these things.
If you want to look at a real culprit in crossovers and amps which is very microphonic and capable of high distortion and hysteresis then look at inductors instead. They are far more reactive in this sense and it is critical to choose the correct design and specification for circuits to limit the effects of microphonics, EM coupling fields, hysteresis and distortion. They are deserving of far more time and consideration certainly when it comes to crossovers.
