On the subject of verification or validity (and on the subject of transfer functions providing a reference too), I copy here a response I supplied to another thread:
Just to clarify a few points where I may not have been sufficiently clear...
Learning occurs all the time in all that we do - at least all of the things that we are aware that we do. Purposeful learning (training) is just where we have some emotional driver to direct the process and highlight the relevant details. A skilled listener may not then need purposeful training, just the relevant experience.
When something is learned, it is generally not unlearned: It is a very non-linear process and makes our full hearing capabilities very difficult to model.
In forming our perception, there exists a substantial capacity for delusion - such as hearing a detail that is not actually sensed.
It is often stated that blind testing can identify delusions in listening evaluations, and I suspect for the most part this is actually true.
However, our inherent non-linearity also means that our hearing capability now can be highly dependent on various emotional drivers and what we might have learned previously: Thus self-proclaimed "golden ears" are afforded a valid defense too.
Hence perception and sensation are not the same thing, and the task of the audio engineer is restricted to optimizing the fidelity of the sensation.
Thus we have the possibility to validate or verify the accuracy of the transfer function where we start with a real acoustic recording. We can also measure errors in that transfer function and correlate them with subjective impressions (via existing measures or by making better use of the measured information we routinely discard). But once we get to individual perceptions and the emotional states that cause their variation, matters cannot be so easily settled.
I think you are confusing sensation with perception. I think this thread also supplies good reason why we might find/already have measurements that can establish a "point of reference" for sensation - and the likely subjective consequences for particular deviations from that reference. IMHO limiting those deviations is the target of high fidelity recording/loudspeaker engineering. You might perceive something entirely different.
So in the spirit of trying to learn from each other as a novel approach….let’s examine your example…….the engineer placing a microphone to capture an acoustic input. The engineer is tasked with choosing the microphone and location based on what he believes are the best tools for this use case…..and in that, a standard of reference beyond subjectivity can never be established.…….this was all choice or free will to fulfill the engineer’s subjective need to capture that input as he or she perceives it. I‘m open to your counter in the hopes of learning something….I’ve been that engineer in the industry for nearly 40 years but again….I’m always open to new or novel ideas.
Firstly, I am here precisely because I wish to learn, and often do. As such, I do not agree that discussion on this forum or any like approach can rightfully be described as "novel".
Secondly, whatever the engineer records is the system input. The task remains then to best engineer the transfer function so that the listener hears that which the engineer heard with the maximum possible fidelity.
(One small proviso that has been reflected in this thread already, is there remains, via the Sound Field or like microphones and higher order recording means, the ability to trade "detail/clarity" for "envelopment/spaciousness" where the listener will intentionally experience a different sound field to the engineer, but with no loss in fidelity. Here I suspect we are short of terminology rather than contradicting ourselves).
I repeat here also that I have been present in the auditoria for several recordings over the years - sometimes in very good seats too - but that on every occasion, listening to what the engineer had recorded showed me clearly that they had the best seat in the house - even when they were often tucked away in a tiny room somewhere else! I learned to regard these recording engineers as experts and to trust that what was presented to me on a disc thereafter was a valid input.
I don't understand how you make 'waterfalls etc' second order and bisprectra third order. Certainly Wiki and the DSP pages I've pre10ded to read don't make things any clearer.
Can you post some bispectra measurements of speakers and explain how they are done?
So there’s nothing new, unusual or interesting to be gained on a forum?……so why engage in a discussion?…….i‘m completely confused by your statement.
From this argument, the listening experience inside an anechoic must be super.
The BBC tried this in the early 70s including with electrostats. The sound was terrible. Much worse than in a listening room. And stereo was terrible too.
It only started sounding OK when they put a floor down in the anechoic
Which anechoic did you use?
BTW, I replicated the experiment in our own anechoic and was surprised at the result ... before Peter Fryer brought the BBC work to my attention.
I'm very suspicious of this. Good recording engineers can set playback level to within 1dB of the original of a live recording with eg Blumlein. If the level is too soft, it doesn't sound further away or the musicians playing softly. It sounds the WRONG SIZE.
Actually I HAVE but it's obvious no one has bothered to read the papers I've linked to
I'll come back to verification and validation in a bit but I want to understand mikets42 & soundbloke's supa dupa measuring methods first.
Hi,
Good info. Can you give me links to any measurements of this specific driver?
I couldn't find anything.
Energy spectra (whether simple spectra or waterfall displays) are a second order measure. Energy is derived from a (non-linear) squaring operation - the squaring being the source of the second order naming. (It also means that our linear measures are derived via a non-linear operation which is confusing too!!).
As you are well aware, these energy spectra can be represented in time or frequency, but are second order measures in either domain - just different versions of the exact same data.
So when we move up to a third order measure, the spectra (very confusingly) become bispectra.
And if we went to fourth order measures, we have "trispectra" - although (thankfully) there turns out to be very little information in such higher order spectra of which we can make use.
Starting from a second order measure then, we need a further convolution (or multiplication in the frequency domain). And in such a measure we have information that pertains to the phase relationships between various frequency components that we lost in making the second order measure.
By way of trying to explain with a real world example... We could hear two sources of a frequency F, one that was a harmonic of one instrument, and a second that was the fundamental of another. The energy spectra do not have information to discriminate the sources of F from one and other, and to cause us the perception of two different instruments where we can identify both via their sonic signature.
The bispectra supply the degree of correlation according to the phase relationship of a fundamental and its harmonics, hence the ability to discern one frequency source from another, or to ascribe a harmonic as "belonging" to its fundamental of one instrument rather than the fundamental of another - especially if we consider the cross-bispectra obtained from having two measures at our ears.
Displaying bispectra measurements is where significant problems are to be found, however. A bispectra has two frequency axes, so requires a 3D plot to start with. A waterfall bispectral plot is therefore very hard to envisage indeed. It is my belief some pre-processing will be required to make the data useful for visual inspecting loudspeaker measures, for example.
However, the domain of speech synthesis and voice identification makes frequent use of the bispectrum (where we can derive data objects that enable us to label a speaker from the relationship of the formants in a speech measurement, for example). I would direct you to this subject to view for yourself - and to see if you can envisage a more appealing way to display the results usefully here too?
Last edited:
I am at a loss to understand your interpretation of my words. There is everything to be gained from learning information here; Learning from a forum or any discussion is not a novel approach. I would politely suggest trying simply to understand the facts that people are trying to communicate, rather than worrying about what any contributor is thinking or is motivated by.
I have read them, and done lots of listening tests. But you might also want to read my comments in post #321 re blind testing too...
Hi abstract,
Again, your entire premise hinges on "could be, you don't know" .... but we do. So instead of swinging in the dark, let's see you come up with factual statements that do apply and are valid - same for comparisons you're trying to make.
Cool, you did a search and failed to understand that is what we are saying. We don't have to say the exact words you searched for. So, what's your point? I've always verified and validated my work. That is an engineering basic, it is what we base our work on. Others are arguing the opposite. In addition, are you suggesting that the demonstrated fallibility of human observation using their own limited senses affected by mood and environment is somehow repeatable enough to rely on? Really?
That actually happens, I have first hand knowledge and direct experience with teams doing precisely that. Marketing didn't tell them to either. It's part of the process.
No, you haven't got any point at all. Accuracy cannot be defined by a system that cannot be calibrated (we have tried). How would you like to be treated medically with instruments that "felt good", like x-ray dose levels? We developed instrumentation for very good reasons. The basic reason is that humans cannot return consistent results in response to the same stimulus. So ... you're trying to tell me that in this instance we can trust human senses, and also they will correlate between humans? Come on. This has been proved to be a false assumption.
You're attempting to draw conclusions by comparing dissimilar things firstly. The codecs you refer to are correcting for measured issues. They do not exist for other media, only a few digital ones.
Again, your entire premise hinges on "could be, you don't know" .... but we do. So instead of swinging in the dark, let's see you come up with factual statements that do apply and are valid - same for comparisons you're trying to make.
Hi mayhem13,
You appear to be trolling.
Most of us have said measurements matter. We have also said they are incomplete so far and that subjective listening is the other part needed. I have taken great pains to say this in many threads, and this one. So you are not reading what I have said.
Now, given you seem to thing that we need to compensate for what is going on inside people's heads, you also know you cannot and therefore everything is pointless. Therefore you have zero value to add. You don't even understand that the hearing senses we have are like different microphones. Therefore you will get different responses between different microphones, but each individual microphone responds the same to the same air pressure variations. Therefore what mic you use falls out of the equation. For that person, their hearing is a constant. That part of your argument should have been painfully obvious to you.
You appear to be trolling.
Most of us have said measurements matter. We have also said they are incomplete so far and that subjective listening is the other part needed. I have taken great pains to say this in many threads, and this one. So you are not reading what I have said.
You aren't. You are attempting to poke holes in it. My personal feeling is that you have insufficient understanding of the science and measurement capability to comment.
Exactly. I have said very clearly that the only reference point can be the air movement in the vicinity of our sensory organs. If you can do that, you have achieved the goal. Once we sense the stimulation, it's our problem how our ears and brain deal with it. Attempting to correct how each person's ear works is pointless, because they will sense the original sound exactly the same way (by definition), and if they don't, something changed with them. We know the brain will process stimulus depending on whatever else is going on, moods etc ... So attempting to raise that as a factor is simply nonsense.
Now, given you seem to thing that we need to compensate for what is going on inside people's heads, you also know you cannot and therefore everything is pointless. Therefore you have zero value to add. You don't even understand that the hearing senses we have are like different microphones. Therefore you will get different responses between different microphones, but each individual microphone responds the same to the same air pressure variations. Therefore what mic you use falls out of the equation. For that person, their hearing is a constant. That part of your argument should have been painfully obvious to you.
Given that I share the same capacity for delusion as any other, I am not sure my view is of any particular relevance. However, I have witnessed people that I consider were influenced in listening tests by factors other than were under investigation, but I have no proof of that. I have also heard claims of people spending large amounts of money on accessories and generating audible performance gains that I find implausible, but I can only guess they were deluded. So FWIW, my experience is that the blind testing suffices if you are not reliant on selling or promoting superlative or somewhat dubious claims.
From this argument, the listening experience inside an anechoic must be super.
The BBC tried this in the early 70s including with electrostats. The sound was terrible. Much worse than in a listtening room. And stereo was terrible too.
It only started sounding OK when they put a floor down in the anechoic
Hi,
I do not think we can make such far reaching conclusions. People do not live in anechoic chambers so brain can not be used to that.
As an example, we have a sound recording room with treated walls, floor & part of ceiling. All sounds like speech, etc are weird there.
For natural representation echo is necessary which is related to environment(psycho-acoustics). Because of that sound engineers in film industry often record in silent rooms and add echos with special devices or computers(nowadays).
Alos I find that in highly reflective rooms most speaker systems do not sound good. Subjective IMO.
Reflections mess up both fr. response, phase & decay timing so it's not suprising that perceived sound quality is significantly affected.
So we are in agreement then.....there exists no reference to either the input or output of the system nor how it is perceived by the listener.?
Please take the time to read at least my contributions to this thread wherein you will find I have supplied explicit answers to your questions already. You might dispute my contributions, but I trust if so, that you will provide some justification for doing so.