Perceptual Audio Evaluation - Theory, method and application

[tuga]

Interesting piece (albeit rather light) on the subject of objective listening or "peceptual measurement":

Perceptual Audio Evaluation - Theory, method and application
Søren Bech (Bang & Olufsen) & Nick Zacharov (DELTA, SenseLab)

https://pdfs.semanticscholar.org/31a2/68ee699dcbacd54202d0acd466a2513529cc.pdf

 

[tuga]

This work has been published as a £180, 464 page book:

https://www.wiley.com/en-br/Percept...heory,+Method+and+Application-p-9780470869239

Abstract​

As audio and telecommunication technologies develop, there is an increasing need to evaluate the technical and perceptual performance of these innovations.
A growing number of new technologies (e.g. low bit-rate coding) are based on specific properties of the auditory system, which are often highly non-linear.
This means that the auditory quality of such systems cannot be measured by traditional physical measures (such as distortion, frequency response etc.), but only by perceptual evaluations in the form of listening tests.
Perceptual Audio Evaluation provides a comprehensive guide to the many variables that need to be considered before, during and after experiments. Including the selection of the content of the programme material to be reproduced, technical aspects of the production of the programme material, the experimental set-up including calibration, and the statistical planning of the experiment and subsequent analysis of the data.
Perceptual Audio Evaluation: Provides a complete and accessible guide to the motives, theory and practical application of perceptual evaluation of reproduced sound. Discusses all the variables of perceptual evaluation, their control and their possible influence on the results.
Covers in detail all international standards on the topic. Is illustrated throughout with tables, figures and worked solutions.
Perceptual Audio Evaluation will appeal to audio and speech engineers as well as researchers in audio and speech laboratories.
Postgraduate students in engineering or acoustics and undergraduate students studying psychoacoustics, speech audio processing and signal processing will also find this an essential reference.

 

[Curvature]

Good reference for listening tests and arguments about setup and results.

 

[Curvature]

I can see the first entry being misused as an apologia by subjectivists. It uses non-scientific or non-reference words as evaluation criteria. I cannot find either explanations of, or references to specific auditory qualities that "cannot be measured by traditional physical measures... "

I need to review it more closely, but it seems to be pseudo-science as far as possible use for music files. As for speech, what can it add that has not been discovered by Bell Labs almost a hundred years ago?

Jim Taylor

That's a little narrow, IMO. A lot has happened.

Here's a quote.

"A growing number of new technologies (e.g. low bit-rate coding) are based on specific properties of the auditory system, which are often highly non-linear. This means that the auditory quality of such systems cannot be measured by traditional physical measures (such as distortion, frequency response etc.), but only by perceptual evaluations in the form of listening tests."

 

[DonR]

ATRAC famously used masking and non-linear techniques to hide low bit-rate codecs, with a mixed reception in its initial implementation. A case of might not measure well but sounds good. The problem I have with these techniques is that they try to fit a standard model for acoustic perception but I think there is a wide variation amongst humans in that regard.

 

[krabapple]

That's a little narrow, IMO. A lot has happened.

Here's a quote.

"A growing number of new technologies (e.g. low bit-rate coding) are based on specific properties of the auditory system, which are often highly non-linear. This means that the auditory quality of such systems cannot be measured by traditional physical measures (such as distortion, frequency response etc.), but only by perceptual evaluations in the form of listening tests."

Yes, the sound of lossy codecs, which rely on psychoacoustic models of hearing, can't necessarily be predicted from things like bitrate or null tests. They have to be evaluated using DBTs. This has been known since forever. even though ignorant audiophile assume 'lossy' means they can always hear some degradation and even though some dishonorable 'demonstrations' involving difference audio were made by figures who should have known better , like George Massenburg.

 

[amirm]

Perceptual codecs have an engine (algorithm) which examines the content and acts differently from moment to moment. The behavior of the system is highly dynamic and varies with content. In addition, depending on the codec, there is psychoacoustic knowledge of our hearing system which it uses to adapt its algorithm for least amount of impairments.

None of the above is true of audio electronics or transducers. They are "dumb" and do what they are told over and over again. For this reason, there is no analogy in evaluating them vs perceptual codecs.

Different systems, different evaluation systems.

FYI I did test our codec at Microsoft using frequency response test and found errors that the team fixed. MP3 encoder from FHG likewise starts to roll off above 17 kHz. So such measurements are useful there but play a very tiny part. Controlled testing is everything.

So don't run with this book and title and think it applies to simple hardware evaluation of audio.

 

[Curvature]

ATRAC famously used masking and non-linear techniques to hide low bit-rate codecs, with a mixed reception in its initial implementation. A case of might not measure well but sounds good. The problem I have with these techniques is that they try to fit a standard model for acoustic perception but I think there is a wide variation amongst humans in that regard.

A much better way to say what I said poorly.

Jim

It's a fair comment and has been addressed by the authors. Wide ranges in listener sensitivities can be controlled through listener selection. They distinguished gradations of listeners, from naive to expert in some specific area. In that way the relevance of tests is established.