Measurements verses listening impressions
Don't you ever wonder why some horns measure flat but sound bad? I have a hunch that Geddes is onto something. Lot's of subjective evidence out there indicating that it makes a big difference on bad horns and noticeable improvements on good ones.
Every time I measure a horn that
sounds bad - and I mean
every time - there is an obvious anomaly, easily measured.
When you see response charts that are smoothed, they hide a world of nasties. So you can definitely hear stuff that doesn't show on a smoothed chart. But an unsmoothed chart is a different story. It shows every tiny little detail, and you can see it in either time domain or frequency domain representations.
Most horns have ripples in response, and almost all show the first couple of quarter wave resonance nodes at the bottom end. Ripples and spikes above that show reflections, diffraction and other anomalies like that. The reactive nature of the horn interacts with passive components in the crossover, sometimes that makes peaks too. All these things are very visible in high-resolution response measurements of horns, truly all speakers for that matter.
I have found that I can
easily hear things that don't show up in a response chart smoothed 1/3 octave. Anomalies that are smoothed with 1/6 octave resolution are also audible. But when I look at a chart that has not been post processed at all, the machine is able to detect subtle details that are completely inaudible, way below the threshold of human detectability. I'm not sure where the threshold resolution is, but I'm guessing it is probably around 1/12th octave. That's about a one note resolution on the chromatic scale. I know I can hear anomalies that are too fine to show up in a chart made at 1/6th octave resolution, but some tiny little details that show up in an unsmoothed chart are inaudible.
High order modes are non-axial sounds that reflect off a boundary, making an early reflection that is delayed in time. Reflections of this sort show up directly in the time domain and they cause self-interference that manifest themselves in the frequency domain. You can see delayed signals in the step and impulse response, self-interference from the same cause blips in the frequency response. But these are pretty small anomalies, seen only the unsmoothed charts of a system measured with relatively high resolution.
Why then, would anyone seeking to show performance improvements in these areas use smoothed response charts as evidence? Smoothing removes all the details one would hope to see. I'm not saying smoothed response charts are necessarily bad - they show an average sound distribution, a spectral balance. That's meaningful. It reduces noise and shows an overall trend. But if you're trying to demonstrate subtle details, reduction of non-axial reflections and edge diffraction, the best way to see them is with a high resolution chart. In truth, probably the
only way to see these things is with high resolution measurements.
So that brings me back to the point. I don't think it is difficult to hear anomalies that are completely invisible in a measurement that has been smoothed to 1/3 or even 1/6 octave resolution. But that doesn't mean you can hear things you can't measure. Maybe you can, maybe you can't. The first thing to do is to measure with high enough resolution, to examine the DUT with adequately high resolution rather than smoothed to 1/3 octave. Comparison with other devices should also be done at equally high resolution. I know for me, there has never been an audible anomaly I couldn't easily find with a good measurement system. The key is to look at the details rather than to mask them out with smoothing.