The correlation between what measures well on test equipment and what sounds good isn’t always as close as perhaps it ought to be, especially when it comes to distortion. Some ‘flavours’ of distortion, such as non-linearity in valves or transformers, can be tolerated in relatively high amounts and in some cases actually contribute to musicality, while others, such as crossover distortion in Class-B amplifiers or aliasing artifacts caused by ignoring the findings of that nice Mr Nyquist, sound offensive even when present at extremely low levels. Similarly, quite high distortion levels in loudspeakers can go completely unnoticed — even when they are hundreds of times greater in magnitude than the level of crossover distortion in their amplifiers.
This raises a problem when it comes to evaluating loudspeakers, as test equipment isn’t really all that good at discriminating between what the human hearing system can tolerate and what it finds offensive. It just spits out numbers. An experienced engineer can go some way towards interpreting the numbers, but once the main parameters are adequately within spec, the only way to make the final judgement is to listen.
Human speech makes a great test signal as it’s something that our brains dedicate a lot of time to processing, so they are good at telling whether it sounds natural or not. If you can find a loudspeaker that can fool you into thinking a real person has just spoken to you, then that’s a good sign that the speaker is doing its job. Sadly most of the speakers I’ve heard have failed this test.
Of course, part of the problem may be that the recording of the voice used to test the speaker wasn’t good enough in the first place, and given the very high accuracy of modern converters and the low distortion of well-designed analogue preamp circuitry, that leaves the microphone as the largest source of variation. Like the loudspeaker, the microphone is electro-mechanical, so after all the numbers have been crunched, you still have to listen to it to evaluate it. The obvious problem is that to listen to the result you need either a loudspeaker or headphones, both complex electro-mechanical devices.
So, we end up in the situation where we judge how natural a loudspeaker sounds by listening to a voice recording made using a microphone while the quality of the microphone is verified by how accurate or musical it sounds over a loudspeaker. And that’s before we even get onto the subject of how the off-axis responses of both the mic and speaker affect the sound in different spaces. The speaker vouches for the mic and the mic vouches for the speaker. Clearly that is far from ideal, but at least it gives the gear geek forums something to argue about.