I measure what I listen to. I ended up experimenting with acoustic foam placement and seeing how it affects the sound field at the listening point(s).
Yes, absolutely, on f < 500Hz everything depends on the mic position and mic type. Our ears are not omni (at all). So you need to find a mic array configuration that resembles your subjective hearing, to the best of your subjective judgment.
Yes, absolutely, on f < 500Hz everything depends on the mic position and mic type. Our ears are not omni (at all). So you need to find a mic array configuration that resembles your subjective hearing, to the best of your subjective judgment.
The Theory of System Identification gives you a slightly different RIR, including on lower frequencies, like:
It also gives you the estimates of RIR estimation errors (matched filters give you none). Everything left of t=0 is the model noise. Thus, you know that what is at t>0 is not perfect, and how badly not perfect. You know then that there is something wrong at f>16kHz (which is, I guess, Barkhausen noise) while the matched filters fool you into unrealistic belief.
The measurement of ... anything ... is not 'X' but 'X+-delta'.
It also gives you the estimates of RIR estimation errors (matched filters give you none). Everything left of t=0 is the model noise. Thus, you know that what is at t>0 is not perfect, and how badly not perfect. You know then that there is something wrong at f>16kHz (which is, I guess, Barkhausen noise) while the matched filters fool you into unrealistic belief.
The measurement of ... anything ... is not 'X' but 'X+-delta'.
or add a suitable waveguide to the microphone that resembles the human outer ear.....and then re correlate everything you thought you knew about measurement spectra.....or not.....just listen and enjoy instead.
Thanks for this Mike. Anyone here willing to do a waterfall from the IR?
My question is, "Do you recognise Angelo's method (it's not just his log chirp) as a Matched Filter measurement?". Thanks for the history lesson.
If you want an error estimate, just do the measurement several times and you will have enough info to do statistical analysis & error estimates very quickly.
What methods do you propose that have better LF noise immunity than Angelo's method within a limited time window?
How do you do your IRs & RIRs?
Last edited:
Can you explain why the noise at t<0 has disappeared by t = 0.4?
Are you claiming this "TSI measurement" gives more useful info than your #434 spectrogram? BTW, 2ms is about 2 feet
============================
I need to do more listening to the residuals that you posted earlier but I'm suspicious of loadsa stuff, especially the crackly stuff which some have claimed is Barkhausen noise.
Someone claimed they can hear Barkhausen when they sweep some units. Still waiting for more details
Last edited:
1. Of course
2. It is not that easy. Chirp does not cover the entire space of excitations, it overfits the noise, and may introduce systematic bias, MLS is even worse. I would get much cleaner results running 1/f noise for the same time as several repetitions of chirp.
3. ReLS on 1/f Gaussian noise of the same RMS as chirp. Sine sweep has 1/sqrt(f) spectrogram and it hurts LF.
4. I use 3 methods: Chirp, Gaussian noise shaped as "average" music which I estimated averaging spectrograms over The Best of Queen CD, and ReLS, and real music of various genres. They all produce slightly different versions of RIR and give enough food for thought. Using sine sweep for tweeters turned out to be a particularly bad idea.
Very good question!
Yes, it is a feature of ReLS which is a biased estimator.
No, I am not claiming anything. I am primarily looking for feedback for everybody errs (including me, of course)
You keep repeating it ...I read the same things in the JC blowtorch thread, it's the same here. Once you have defined what is sound and we are on the same plane (?) then you can start to use the term and conjugate it.
But I don't blame you for not catching it, it's the times.
Sorry, but I have trouble folllowing this thread. The title is about driver measurements, but discussion is all around room sound, mics, analysis methods and presentation graphics, loudspeaker directivity, listening test preferences etc. Some interesting points from history though
Hello Juhazi,
You are not alone: I have the same trouble. Looking back it seems that - between the lines- there is a mix up between (an improved way of?) measuring distortion and the audibility of distortion. Then the thread meanders on and off towards better or worse sounding loudspeaker systems.
History repeats itself in this thread: lots of belief in Waterfalls, Wigners, Wavelets and Novel Distortion measuring methods as an explanation for "better" sounding systems.
Alas, Toole and Olive have shown matters to be otherwise, not matter the methodological shortcomings of their ventures.
I think this is what kgrlee tries to explain here: been there, done all that..
You are not alone: I have the same trouble. Looking back it seems that - between the lines- there is a mix up between (an improved way of?) measuring distortion and the audibility of distortion. Then the thread meanders on and off towards better or worse sounding loudspeaker systems.
History repeats itself in this thread: lots of belief in Waterfalls, Wigners, Wavelets and Novel Distortion measuring methods as an explanation for "better" sounding systems.
Alas, Toole and Olive have shown matters to be otherwise, not matter the methodological shortcomings of their ventures.
I think this is what kgrlee tries to explain here: been there, done all that..
Sorry, but I have trouble folllowing this thread. The title is about driver measurements, but discussion is all around room sound, mics, analysis methods and presentation graphics, loudspeaker directivity, listening test preferences etc. Some interesting points from history though
If it had all been done before, then there would no questions posed in the thread opener that did not have answers already.
New thread - with a bit different focus...
https://www.diyaudio.com/community/...needed-for-speaker-design.410219/post-7625237
But this thread does not contain discussion of any "new distortions", nor any "new measurements", nor actually any "new means of presenting distortion". What you have done in your new thread is constrained the discussion to linear measures and lumped parameters - all of which is well-known, well documented, and provides for a well tried-and-tested means of designing loudspeakers.
I have refrained from posting this on the new thread to avoid being characterized by negativity and littering, but I fail to see the new thread asks anything that is not already very well known. What is to discuss? What am I missing?
Further to my last post, I was going to suggest the only variable in the new thread (if we excuse passive crossover design which is a whole new topic) is the choice of Qts. For what is IMHO the best read on the subject of Q, I was also going to simply provide a reference to Michael Gerzon's paper "Why Do Equalisers Sound Different?" (published in Studio Sound, July 1990), and leave it at that.
However, since it has been over 30 years since I read the article properly, I thought I would have a read once more. The following statement stood out starkly...
"the Wigner distribution may well form the basis of future improved methods of time/frequency analysis beating the uncertainty principle limit. This is because it can be shown mathematically (the methods of proof are buried deep in the Quantum Theory literature) that the normal methods of time/frequency analysis can be obtained from the Wigner distribution simply by blurring it with a suitable smoothing filter."
He continues...
"The point is that real-world measurements are never exact, and what we shall see is that incredibly small changes in amplitude and phase response, supposedly quite ‘negligible’ according to objectivist ideas, can actually have large audible effects. This is not to say that these effects cannot be measured, only that measurements of amplitude and phase responses are not the way to do the measurement."
"we shall have to stop thinking of filters purely in terms of their amplitude and phase responses but will have to find other new ways of looking at them."
Perhaps every contributor/reader here would do well to read this article too? I trust no one here is up to challenging the intellectual might of MIchael Gerzon? I'm certainly not!!
"Can" in this sentence refers to the application. The differences between the detection in pre- and post-echo is one obvious example that is backed up by plenty of research that I hope you have not missed.
And just for the record, I learned long ago never to trust my own opinion. On the rare occasion I have described a subjective impression, I have stated it clearly as my own, and offered it as an alternative to that which has been reported by others whose abilities I learned to trust well.
Perhaps true for some experienced designers. There are a lot of people just getting started and might benefit from a "everything measurements in one place" thread.
I consider the process of making accurate, repeatable measurements to be one of the most challenging aspects of DIY speaker design. The fact that it is so often done wrong, even by experienced people, makes my case. I don't think I have ever done a complete set of frequency response measurements without needing to repeat part of it.
So there is plenty of "discussion headroom" to talk about tools, techniques, methods for QC and data reduction...