Amateur speaker designer/builder, DSP programmer.
System: my own three-way speakers with linear phase, DSP active crossovers, driver correction, time alignment, calculated in-room EQ to compensate for directivity characteristics. I wrote the DSP software in 'C' using the FFTW library to perform the filtering and convolution; it runs on Linux.
I seem to be the only person around here who is sceptical of all of the following:
- in-room measurements
- target frequency responses
- room correction
- listening tests
- the statistics used in audio experiments
- dipole speakers
- bass reflex
- frequency domain-based thinking
What does that leave? Well, it leaves speakers and systems that are purposely designed to behave
very simply - not that the hardware or software is necessarily simple.
Such systems are not contrived kludges based on feedback from over-simplified in-room measurements that assume no ability for human hearing to interpret the audio 'scene', nor from pseudoscientific experiments and the torturing of statistics from inconclusive listening tests; they don't have super-narrow, unnatural, radically modified dispersion; they don't spray antiphase around the room; they don't make assumptions on how human hearing works based on an analogy with a microphone and Fourier transform; they attempt to reproduce the time domain
waveform and don't justify phase and timing distortion on the basis of 'legacy' hardware, listening tests and statistics; they implicitly assume that evolution logically gives a human the ability to separate direct from reverberant sound, and therefore logically
room correction is not valid.
The common ASR response to all this would be "Prove it..." or, if then pinned down on the true definition of science, "Demonstrate its likely probability..." or some such. But in doing so, the would-be scientist strays into a minefield of problems that can, to be sure, be waved away with some stated assumptions, but which means that everything derived from the experiments is based on assumptions that have no
I could create a large list, but let's start with "It is assumed that a person's awareness they are taking part in an experiment does not alter what they hear nor hinder their ability to discriminate...". Not many people would even bother to state this assumption, but it is crucial to any scheme that seeks to justify the design of an audio system using feedback from listening test results. Everything that follows from every listening test is based on that assumption, but without a Truman Show type setup, we can never know whether it is valid or not. Our own inner experience, however, suggests that it is not...
Whether that 'inner experience' is real or an illusion based on expectation bias, confirmation bias or any other bias, we cannot know. So is it safer to base audio system design on listening tests, or is it safer to make rational assumptions that bypass both listening tests and subjective 'inner experience'? For example "The audio system should reproduce the time domain waveform without error".
This apparently-sensible ideal is not too far from being achievable these days, but the audiophile world is so deeply under the spell of frequency domain thinking that it cannot actually understand what the words mean, never mind aspire to it. Thus, a multi-way speaker that aligns inter-driver phase through its crossovers is regarded as perfect because that results in no overall frequency response magnitude dip, but any further moves towards genuine fidelity to the signal (linear phase and genuine time alignment between drivers using delays) are regarded as the obsessions of a crank.
Another sensible idea based on logic is: "Human beings can identify the nature and location of acoustic sources in a reverberant environment; therefore they are separating direct from reverberant sound; therefore in-room frequency response measurements do not represent what we hear, and therefore 'room correction' is meaningless". Again, frequency domain thinking and the necessity to represent everything as a simple graph on a laptop screen ("It needs smoothing of course") means that these words are literally not understood.
In audio, it is better to design things based on feedforward
logic and justified assumptions, rather than experimental feedback
which is, ultimately, based on unjustified assumptions.