You could use Noise in short bursts, Average linger time(1) of a note is ~.75-1.5 seconds (2) so that might allow enough time to move the mic while still limiting noise induced error.
With sweeps you would need to do incremental movements and lots of samples.
This is an interesting idea, although it leaves out head shading of the ear opposite of the reflected sound, thus reflection energy will be measured louder than a person would perceive them.
edit: (1) average linger time on the recording not in room.
(2) at 80bpm (slow song)
With sweeps you would need to do incremental movements and lots of samples.
This is an interesting idea, although it leaves out head shading of the ear opposite of the reflected sound, thus reflection energy will be measured louder than a person would perceive them.
edit: (1) average linger time on the recording not in room.
(2) at 80bpm (slow song)
Last edited:
I have to say, after doing some quick tests in my calibrated space (which now that I think about it is quite easy to measure... huh) That this technique seems to work quite well with pulsed pink noise. 1Meter/1Second, took three samples to cover sound field at head height sitting (listening position) Next time I'm somewhere more difficult I will have another go.
I like
I took too many samples per second and my poor computer is still crunching numbers... but the average seems to have settled. Should be done sometime this week.
I like
I took too many samples per second and my poor computer is still crunching numbers... but the average seems to have settled. Should be done sometime this week.
That's almost a whole different subject. One akin to HRTF and the target slope. I find that the 1dB/octave downward slope helps compensate for the off-ear response a microphone doesn't hear.
HRTF is exactly what it is... how much it matters
. I agree it's almost a different topic, almost. After giving up on sleeping more than 4 hours tonight, after dreaming about microphone shielding techniques to measure direct sound vs reflected sound using the MMM technique I thought i'd post my quandary. The MMM is good at gathering an average sound profile of the space measured. But Microphones do not "hear" the same way we do and can detect interference that we do not above 1,000hz. With enough samples this frequency interference should be averaged out but what is a large enough sample size to do this and produce useful information.
A tricky balance, much like all of audio I suppose.
The wife just woke up, time to make some breakfast.
It seems that the moving mic is a good technique because it takes so many averages over the area in which you move the mic. It has the advantage of being a fast way to get an average response. One of the papers that Jean-Luc linked to shows the correlation of different averaging techniques. It's worth reading.
The MMM does not turn the microphone into an ear on a human head. It just gives a more complete picture than a single point measurement.
Back in the old days, I used to do Dolby and THX alignments in cinemas. Those standards required measuring at several points across the seating area. 5 to 7 points was typical, IIRC. That wasn't really to get combined average, but to be sure that everyone got to hear the soundtrack within spec.
In P.A. work we are supposed to measure at several points, but rarely do. It will be interesting to see if the MMM can be adapted to large spaces.
The MMM does not turn the microphone into an ear on a human head. It just gives a more complete picture than a single point measurement.
Back in the old days, I used to do Dolby and THX alignments in cinemas. Those standards required measuring at several points across the seating area. 5 to 7 points was typical, IIRC. That wasn't really to get combined average, but to be sure that everyone got to hear the soundtrack within spec.
In P.A. work we are supposed to measure at several points, but rarely do. It will be interesting to see if the MMM can be adapted to large spaces.
I worked with some of the early Meyer rigs and we would set mics all over the auditoriums depending on size of course, 12 was average. Useful for setting the side delays and beam-forming the front end. Final EQ was almost always set with a CD of the show album by ear as the Auto EQ could get a bit wonky. That was a long time ago.
Sorry for the OT
I'll look for the white paper you mentioned.
Sorry for the OT
I'll look for the white paper you mentioned.
One thing that people must understand about this topic is that it only applies to the steady state measurement of SPL in a room. One cannot look at it as saying anything about direct to reflected ratios or head shadowing or the response to non-stationary signals like music. Things will just get all mixed up if we don't keep within the limitations of the topics assumptions.
As I explained in the MMM note http://www.ohl.to/audio/downloads/MM...easurement.pdf, the averaging lessens the influence of reflections because the reflections are uncorrelated, contrary to the direct sound.
This topic is about a reliable method to measure the frequency response of loudspeakers in a room, with a good enough (?) precision. Not to measure SPL only.
In draft standard ISO 16283-1 (see http://www.simmons.se/Filer/PDF-filer/AkuLite-no-3-Uncertainties-Simmons-SP-report-2012-28.pdf, pages 16 and after), you can see third-octave measurements with differents methods : this shows that the manual microphone scanning is used for frequency curves not only SPL.
I'm afraid I don't understand what you mean about non-stationary signals : MLS, sinesweep, are non stationary but you get the true impulse response at one position for an LTI system. And so you know what will happend to any signal including music. Am I wrong ?
Averaging with pink noise, you loose the phase but you get the mean amplitude response.
Last edited:
jlo
To me SPL and frequency response are the same thing, but you may be thinking of SPL as a single number measure of level. I meant it as SPL(f). (But you are right, my usage here is probably sloppy.)
The statistics from Schroeder are only valid for a steady state measure. They clearly show that any single point is not very stable and some averaging must be done. If you want to know how much averaging is required then we can only talk about this if we are talking about the steady state. The steady state only has meaning in the frequency domain. Hence, we cannot really talk about the steady state of an impulse response. We can take the FFT of this impulse to get the steady state, but only as long as the FFT length covers the whole impulse response.
What I am saying is that this topic is complex and we have to be careful about what we mean (although I guess I wasn't!).
If you use noise as a signal then it is stationary and you should get the same answer if you use a sweep as long as the frequency response derived from this impulse uses a long enough time. But at this point we are still only talking about single points. Doing a sweep measurement with a moving mic adds in so much theoretical complexity that I would not even begin to try and analyze that.
Basically I am in complete agreement with you on all points. (Except that MLS is stationary if the period is > RT60). It is the discussions that mention head shadowing and the like that I see no basis for because these things are time signal dependent and cannot really be talked about in a steady state sense.
The use of a moving mic for spatial averaging is many decades old and there is a lot of information out there on this topic. It is certainly a convenient way to do spatial averaging for small room FRF measurements, but it does not provide any valid time domain information.
If one wanted to combine spatial averaging with psychoacoustic properties, then one would need to look at the ears integration time and match this to the averaging time of the spatial averages. You could not do that with a moving mic, but it can be done with stationary points. I have done this before and implemented techniques for automotive systems that use this technique, but that is proprietary (done under a consulting agreement) and not something that I can discuss in detail in an open forum.
To me SPL and frequency response are the same thing, but you may be thinking of SPL as a single number measure of level. I meant it as SPL(f). (But you are right, my usage here is probably sloppy.)
The statistics from Schroeder are only valid for a steady state measure. They clearly show that any single point is not very stable and some averaging must be done. If you want to know how much averaging is required then we can only talk about this if we are talking about the steady state. The steady state only has meaning in the frequency domain. Hence, we cannot really talk about the steady state of an impulse response. We can take the FFT of this impulse to get the steady state, but only as long as the FFT length covers the whole impulse response.
What I am saying is that this topic is complex and we have to be careful about what we mean (although I guess I wasn't!).
If you use noise as a signal then it is stationary and you should get the same answer if you use a sweep as long as the frequency response derived from this impulse uses a long enough time. But at this point we are still only talking about single points. Doing a sweep measurement with a moving mic adds in so much theoretical complexity that I would not even begin to try and analyze that.
Basically I am in complete agreement with you on all points. (Except that MLS is stationary if the period is > RT60). It is the discussions that mention head shadowing and the like that I see no basis for because these things are time signal dependent and cannot really be talked about in a steady state sense.
The use of a moving mic for spatial averaging is many decades old and there is a lot of information out there on this topic. It is certainly a convenient way to do spatial averaging for small room FRF measurements, but it does not provide any valid time domain information.
If one wanted to combine spatial averaging with psychoacoustic properties, then one would need to look at the ears integration time and match this to the averaging time of the spatial averages. You could not do that with a moving mic, but it can be done with stationary points. I have done this before and implemented techniques for automotive systems that use this technique, but that is proprietary (done under a consulting agreement) and not something that I can discuss in detail in an open forum.
Last edited:
OK, then I understand and agree.
I also completly agree with this.
Maybe there is a way to do it with a moving mic ? But basically, we know it works with stationary points. We can do "psychoacoustic windowing" on many impulse responses and then average. Or we can even average impulse responses (with some caution and tricks) and after do some psychoacoustic signal process. This may be a next step.
But I think that the magnitude frequency response is still the main parameter for the best listening quality and Moving Mic Measurement method can be a convenient tool to achieve this.
Well I wanted to get phase response at listening place. Normally when you analyze the phase of an impulse response in a room, with all reflections, it is difficult to get something readable and coherent. So I thought about averaging phase response but this cannot be done directly. Why not average impulse response and get an average magnitude and phase ?
Having never heard of averaging impulse responses, I tried it and was surprised that it worked. You cannot do it directly, there are some tricks, but at the end you get an averaged magnitude and averaged phase.
I checked if results were the same :
- impulse responses > fft > magnitude > average of magnitudes
- impulse response > average of impulses > fft > magnitude
As I got nearly same results from both methods, I'd say that impulse averaging works and I finally got the averaged phase response at listening place.
jlo
Yes, quite interesting, but not really my question since I am sure that your technique of impulse averaging would work at individual points.
My question is this. If I run HolmImpulse using a log sweep and a moving mic will the resulting impulse response be reasonable? I don't see how it could be, but you never know, weirder things have happened.
Of course running Holm at various points - with mic stationary - will yield average-able impulse responses just as you suggest. And one can apply some psychoacoustics to the impulse and the averaging - for example higher frequencies might want to de-weight more distant points from the listener. The LFs will want all points with a long time average, while HFs will want closer points and a shorter average.
If a swept Holm does work, then what kind of average would this be? Time as well as space?
Yes, quite interesting, but not really my question since I am sure that your technique of impulse averaging would work at individual points.
My question is this. If I run HolmImpulse using a log sweep and a moving mic will the resulting impulse response be reasonable? I don't see how it could be, but you never know, weirder things have happened.
Of course running Holm at various points - with mic stationary - will yield average-able impulse responses just as you suggest. And one can apply some psychoacoustics to the impulse and the averaging - for example higher frequencies might want to de-weight more distant points from the listener. The LFs will want all points with a long time average, while HFs will want closer points and a shorter average.
If a swept Holm does work, then what kind of average would this be? Time as well as space?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.