Window overlap does insure you have the single window that has the complete spectral peak. However, depending on FFT interval, the amplitude is still distributed over the duration of the FFT window. You can derive a peak value (1.414) from the window, but this is not true peak, but a mathematical construct from the average.
You hit the nail on the head RobHello Camplo
Look up M noise as far real world requirements for PA applications headroom wise. The average is arbitrary depending on the circumstances. Clapping and expecting to see it on an RTA in real time??? Why don't you set your sample size higher to average the spectrum and slow down the visual response and just try repeated clapping to capture the spectrum. You should see the peak in the average spectrum.
https://m-noise.org/
Rob
They just reiterated my earlier post but with better and more proper wording lol.....Crest factor generally rises with frequency as displayed within this chart of many examples super imposed upon themselves.
Can anyone interpret the frequency of this graph? Reminds me of how manufacturers use odd metrics to keep people from figuring out how things measure up.
Last edited:
The above graph is a visual representation of the data I sought to develop on my own using a fast acting RTA....I could use this graph to ball park the headroom profile I seek to create and expect of my own system vs the average levels I desire at desired listening distance vs......FREQUENCY.
my guess is that we have intervals at 100hz, 1000hz, and 10,000hz
I'd need to create a curve to superimpose over pink noise displayed at 1 octave scaling to make this information most effective or make sure to stay in the context of 1/3octave RTA
The 115db fork drop is arbitrary outside of how loud a real event can be.... it has no relation to an amount of gain over input signal making it useless to calibrate a system, only stating the dynamic requirement to recreate said event while still void of frequency specificity.
The above graph, if trustworthy, along with an accurate spl reading on an RTA would expose exactly where 115db takes place...likely around 1000hz or higher. Using the above graph putting 115db, 15db over an typical average of 100db.... which is easily 10dbs too loud. I can speculate that a more dynamic source like a movie might still hit a higher crest factor, reaching 115db peak while average will be much lower, since 90db is loud and all parts of movies are not.
I don't need 15db of headroom at 100-500hz...according to the graph above, 10db is more realistic. If aiming for say 95db as an upper monitoring level, which should be generous, then 105db would be transient level for that part of the spectrum.
👍
my guess is that we have intervals at 100hz, 1000hz, and 10,000hz
I'd need to create a curve to superimpose over pink noise displayed at 1 octave scaling to make this information most effective or make sure to stay in the context of 1/3octave RTA
The 115db fork drop is arbitrary outside of how loud a real event can be.... it has no relation to an amount of gain over input signal making it useless to calibrate a system, only stating the dynamic requirement to recreate said event while still void of frequency specificity.
The above graph, if trustworthy, along with an accurate spl reading on an RTA would expose exactly where 115db takes place...likely around 1000hz or higher. Using the above graph putting 115db, 15db over an typical average of 100db.... which is easily 10dbs too loud. I can speculate that a more dynamic source like a movie might still hit a higher crest factor, reaching 115db peak while average will be much lower, since 90db is loud and all parts of movies are not.
I don't need 15db of headroom at 100-500hz...according to the graph above, 10db is more realistic. If aiming for say 95db as an upper monitoring level, which should be generous, then 105db would be transient level for that part of the spectrum.
👍
Last edited:
One could use a pink noise source to voice their system in the shape of the crest factor trend line using a 1/3oct RTA... then take a sine sweep measurement to view thd levels of the predicted transient spl level....
The blue line is an example of the Trend line you might follow to voice your system playing pink noise, for the test...Taken into consideration that pink noise is generally flat on a 1/3oct RTA, you would just adjust the idea for that.
Basically, after setting up voicing as described, run a sine sweep that is approximately 7db above desired listening level, at 100hz.
Another view of the potential test voicing curve...
As long as the idea is understood, one could choose a test voicing slope of their own discretion
The blue line is an example of the Trend line you might follow to voice your system playing pink noise, for the test...Taken into consideration that pink noise is generally flat on a 1/3oct RTA, you would just adjust the idea for that.
Basically, after setting up voicing as described, run a sine sweep that is approximately 7db above desired listening level, at 100hz.
Another view of the potential test voicing curve...
As long as the idea is understood, one could choose a test voicing slope of their own discretion
Last edited:
Marcel will no doubt appreciate your effort to back him up.
Unfortunately your reasoning is wrong, out of context.
To refresh your memory, your pupil said this:
Then he admitted:
So this discussion clearly was not specifically about Camplo's horn, but about a specific excerpt from Gunness' paper.
As a scientist, you should know better than anyone else how important it is to be specific.
I was just thinking...You wouldn't have to go as far as literally voicing the system. It would be what you are doing if your system was already voiced neutral as far as pink noise is concerned. Rather, you might simply apply a filter of this eq'ing on your master channel. Then take the sine measurement. Then you'd be able to test your unique voicing choices with a "predicted headroom curve"....there, I made up another term.
Deja RTA?
check out your post from: https://www.diyaudio.com/community/...ion-with-a-2-way.334757/page-498#post-6938711
and the immediate reply
and more posts both preceding and succeeding the linked one
I don't think you will find a RTA that captures "True Peak".
An RTA is a device that displays energy in frequency bins (or fractional-octave bands) ...that is it's purpose.
It may provide statistics about each bin, such as average and peak, but they are time-blind statistics about the summed energy within the bins.
And the energy in each bin is a function of the bins' width, so a hand clap will somewhat counterintuitively read a higher peaks with the REW RTA Mode set to 'RTA 1/1 octave', than it will at 1/48th octave.
(In Spectrum mode bin width is controlled by FFT size, bins will be small, so an 8k FFT @ 48kHz will have a time window width of 166ms, (ignoring Overlap and windowing). Peak will be an average of that time window as best i can tell.
Bottom line RTA is the wrong tool for capturing "true peak".
Which begs the question...."what is true peak?"
And how fast are our SPL meters?
Iow, what is the time window that is being averaged (weighted) by a peak reading meter like REW peak?
It seems serious SPL meters conform to the IEC 61672-1 standard.
Knowing the REW's technical excellence, i bet it does too (couldn't find for sure in manual)
Slow has a time weighting of 1 sec; Fast 125ms, and Impulse 35ms.
I think REW's peak is highly likely Impulse 35ms.
here is a good article about how SPL meters and the IEC standard work: https://www.nti-audio.com/en/support/know-how/fast-slow-impulse-time-weighting-what-do-they-mean
As far as how fast do we really want meter to be to capture a "True peak"......well, what time windows matters to you, and for what purpose?
Peak limiters attack settings to protect drivers from over excursion , are often less the 35ms even for subs. And less than 1ms for CDs.
But that's for live-where a mic can be dropped etc...and very fast transients can wreak havoc.
For recorded music, as mentioned previously, peaks such as a mic drop are not an issue. (Although be careful with some of Tom Danley's homemade recordings like fireworks etc
Personally, I don't even use or care about a peak meter for anything other than curiosity when thinking about system head room for recorded music.
I say to self, what's the max average SPL Z i think i'll ever want. Then I add 20dB to that, and see if all my drivers have the ability to reach that "peak".
To prove that someday, a dog chasing its tail can actually catch it, i guess hahaAnd this is needed because?
You can download the noise file from the site. All it's used for is showing the actual peak SPL of the system at test. There is nothing to voice too?? It's showing you that pink noise crest factor doesn't equal music crest factor under real world conditions and pink noise is not a suitable test method to determine if the loudspeaker systems can meet the crest factor of an actual live performance.
It also gives you a metric to compare systems with tested under the same conditions.
Rob
The REW logger records the SPL (with the currently selected weighting and time constant), the minimum and maximum values, the unweighted and uncalibrated peak value, the equivalent sound level and the sound exposure level.
LZ (Lpeak) has no time weighting applied, the signal is not passed through an RMS circuit or calculator, it is the maximum value reached by the sound pressure.
LZ is the true peak of the sound pressure wave, assuming the measurement microphone does not clip or limit.
Last edited:
Hi Art, i think it would be a good question for John M...exactly how does the LZ peak meter work.
I would be very surprised if it works like you are saying, a true instantaneous, ie 1 sample pressure.
I think it has to have an integration period. ARTA use 35 ms for LZI-Impulse. Smaart doesn't say in their manual, either, but they do use 1s for Slow, and 125ms for Fast ....like all three programs do.
Last edited: