Objective, repeatable sound quality measurements and metrics

Question

I am currently doing an electronics design for a basic 2.5W speaker amplifier for an embedded system based on COTS parts (e.g. codec and amp), similar to what one would find in a tablet.

Aside from a wide range of parts to choose from, there are other engineering trade-offs, such as the filter on the CLass D amp for EMI/EMC and the noise on the power supply rails.

This, of course, is even before deciding on a speaker.

Is there a standard method for an objective and repeatable evaluation process? If so, what sort of equipment and methods do I need, and what metrics do I need to calculate, and how to these metrics relate to the perception of audio quality?

(Yes, of course, there is SNR THD and frequency response, but I am yet to relate these to how a human perceives the quality of the audio).

(Second note: This isn't an evaluation of codecs, compression, sampling or other DSP artefacts. I have control of the bitstream I send to the audio codec chip. Advice of choosing the signal I should send is most welcome).

Answer 1

i dunno anything about EAQUAL or PEAQ.

judging which sounds better is quite subjective, so much so that i consider it nearly a useless question.

but if you have a reference (that you define as "good"), and you have several options to compare to the reference, and your goal is find which option most sounds like the reference, then i would recommend AB testing (a little bit different from ABX testing). at least blind, and better yet, double blind.

the subject listens to A and to B. as many times as they would like. pressing A always gets them sound A and pressing B always gets them sound B. then the only question to ask the subject is "Do these sound the same or are they different?"

put in an equal number of pairs that are truly identical as those that are the difference you're trying to test. so for two different sounds, the subject hears AA, AB, BA, and BB. if the subject scores more true positives than false positives and more true negatives than false negatives, he or she might very well be hearing the difference. get lotsa subjects and have them hear lotsa different sound pairs and you can get some meaningful results.