The goal of audio reproduction


I have followed the ongoing exchanges regarding subjective versus objective criteria for audio quality for some time now and finally wish to toss in my 2 cents. I am a Cognitive psychologist and have spent most of my professional life conducting research to measure how consumers and users of products perceive those products. I use formal psychological measurement techniques to quantify perceptions. In essence, I "objectify" subjective experience. So, the subjective versus objective distinction is, to me, somewhat misconceived. Given that perceptual experience can be measured (objectified), I would say the important question is, what physical features of products (i.e., physical audio metrics) correlate most accurately with perception? A more valid distinction, to me, is the distinction between "effective" (something of a causal nature) versus "affective" (i.e., something of an emotional nature). Objective product metrics "effect" subjective perceptions (affects). It's stimulus/response psychology.

I use multivariate statistical techniques to model effect/affect (i.e., stimulus/response) relationships. Here, it is important to consider that perception (i.e., affect) is not a single thing, but a composite of multiple factors. These include a) Valence (the standard good versus bad distinction, b) Potency (strong versus delicate feeling), c) Arousal (intense versus mild feeling) and d) Novelty (familiar versus unusual feeling). Hence, a complete description of perception requires a profile of these four perceptual components and not a single concept such as good or bad.

It is also important to note that perceptual data such as described above are obtained from samples of typical product users not from one, or a few, experts. Trained experts may be used in perceptual research, but their role is simply to describe the subtle perceptual qualities of products that may elude measurement. Experts serve as biological test instruments (very common in food, beverage, and cosmetics industries). But they don't serve as surrogates for typical consumers of products. For that, perceptual data are obtained from typical consumers. And those data are not expressed in terms of perceived physical properties of products, but the psychological responses to those physical properties.

The upshot of this for audio (it seems to me) is that if you just want to replicate a particular auditory event (e.g., a musical performance) then matching all physical metrics of the reproduced performance to the live performance is fine. But if you want to produce an enjoyable audio experience you might find that certain physical properties affect listener's perceptions more or less than others and that a departure from veridical presentation is preferred. For my part, I think that is fine. In fact, I am finding my tastes in audio quality are changing somewhat and I am now drawn to a slightly different quality in my system. I don't care if it matches the real performance exactly. I care that I like it.

jakleiss

kota1,

As to segmentation, yes, I have done a good bit of that as well. There are the typical factors such as age, gender, education, and culture. But, people just differ from one another in their individual preferences and sensibilities. Those differences are easily reflected in different profiles of the perceptual scores I mention.

That's a lot of science to come to the conclusion; I like what I like because I like it. 😄 In all seriousness jakleiss that sounds like a very thought out and reasonable approach though I do wonder what the value is, i.e. who your audience would really be for this research.

@Jond The point for me is just that you can't claim a technical metric is more desirable unless you've actually done the testing for desirability.

 

@jond, 

I was a bit loose in my use of the term "like" as that only describes the Valence component of perception. In reality, people in the Sensory Sciences often only use that one component. I like the full perceptual profile, so "Preference" might be a better term. The point, however, is that the science provides knowledge of the physical properties of the system that are most strongly associated with each individual perceptual component. Knowing that enables choosing a product for listening whose metrics are most closely aligned with preferences. Ultimately, you do have to listen, but the science helps you weed out the chaff before listening.

I often begin a study by having the participants rate their imagined "ideal" (i.e., preferred) product. That provides a criterion against which to compare the actual products being evaluated in the research. It often happens that different individual products exemplify ideal levels of different psychological components. Hence, the "ideal" product for a given sample of listeners may not yet exist. The challenge for designers and engineers is to combine those separate features into a single "ideal" product. 

I want to emphasize that the research informs group tendencies. However, the statistical nature of the data allow you go backwards and specify for a given perceptual profile, what physical metrics correspond to it. Doing so gives you a head start on choosing a system for listening, which is the ultimate arbiter of quality for you.