Agreed.
Salmi uses smoothing with about 100 Hz constant bandwidth up to 500 Hz and a little narrower than 1/3rd octave constant percentage bandwidth above. His time window approach is all analog (this is from 1983) and uses a 4 step approximation to an exponential window. Window length is inversely proportional to frequency and it looks like t is approximately 10ms at 1000 Hz. (So 1 ms at 10kHz.)
David
Can you further explain what you would recommend as appropriate? A variable window that hits certain targets and specific smoothing targets across the spectrum?
I'm interested what targets you would recommend at which frequencies.
I would have the time window inversly proportional to frequency up to about 500 Hz, at which point it should be about 10 ms. Then I would hold it constant at 10 ms. I would also use more smoothing at LFs, maybe 1/2 octave or 1/3 octave, maybe even constant bandwidth as Dave commented. Although 100 Hz seems pretty wide since that makes everything below 100 Hz a single number. That seems excessive. I would taper this bandwidth down to 1/6 or 1/10 ocatve above 1 kHz.
I would like to point out that the "ears bandwidth" or ERB as Moore calls it, that is often talked about here, are not developed to represent the frequency resolution of a measurement as we are using it here. There is some validity, but its not a precise relationship. Some frequency averaging is always required when you use an FTT because the resolution at HF is far greater than the ear and the ear has very low resolution at LFs (but so does the FFT). There is a lot of discussion of what is right. Toole holds that 1/20 octave is required, and that could be for some systems to show very sharp resonances, but I think that it might be too narrow for the kind of "rough" data that we are forced to accept when we don't have an anechoiic chamber. The most common is 1/3 octave and I would agree that this is too wide, especially when polar maps are used because there is some smoothing in the angular direction as well and this smoothing on top of 1/3 octave makes more like 1/2 octave or wider. So I think that precise numbers are hard to estimate, except to say that 1/3 octave is too wide and 1/20 octave is not achieveable in many circumstances.
I would like to point out that the "ears bandwidth" or ERB as Moore calls it, that is often talked about here, are not developed to represent the frequency resolution of a measurement as we are using it here. There is some validity, but its not a precise relationship. Some frequency averaging is always required when you use an FTT because the resolution at HF is far greater than the ear and the ear has very low resolution at LFs (but so does the FFT). There is a lot of discussion of what is right. Toole holds that 1/20 octave is required, and that could be for some systems to show very sharp resonances, but I think that it might be too narrow for the kind of "rough" data that we are forced to accept when we don't have an anechoiic chamber. The most common is 1/3 octave and I would agree that this is too wide, especially when polar maps are used because there is some smoothing in the angular direction as well and this smoothing on top of 1/3 octave makes more like 1/2 octave or wider. So I think that precise numbers are hard to estimate, except to say that 1/3 octave is too wide and 1/20 octave is not achieveable in many circumstances.
Holm uses a continuosly varying window width that is linear from LF to HF. Its width is adjustable, but not as to its change with frequency. So invariably you get one end of the spectrum wrong, or both.
but ERB windowing would not be so far of your recommendations :
it is about 1/9th of an octave above 500Hz, levelling to 25Hz at low frequencies
https://ccrma.stanford.edu/~jos/bbt/Equivalent_Rectangular_Bandwidth.html
After windowing, you may smooth the resulting curve with other parameters.
It seems to me that when Toole speaks about fine resolution measurements, it is more to analyse specific problems rather than to balance an in-room frequency response.
Back to smoothing, has somebody tried non-classical methods such as "spectral enveloppe" smoothing (see DRC: Digital Room Correction or maybe "true enveloppe" (see http://articles.ircam.fr/textes/Roebel05a/index.pdf
Just a note about the 5ms window used by Doug Plumb : I don't know with type of window he uses but when measured, his window is nearer to a 9ms Tuckey window. So when we speak of a window length, the window type (Hann, Blackman, Tuckey,....) is also quite important.
Last edited:
Just to touch on a previous topic w/ regards to reflections I found this here:
http://human-factors.arc.nasa.gov/publibrary/Begault_2000_3d_Sound_Multimedia.pdf
Goes right along with the position expressed by Dr. Geddes on the matter. That case is completely closed for me.
Anyway, back to windowing,
Dan
http://human-factors.arc.nasa.gov/publibrary/Begault_2000_3d_Sound_Multimedia.pdf
Goes right along with the position expressed by Dr. Geddes on the matter. That case is completely closed for me.
Anyway, back to windowing,
Dan
I constantly use ERB with gammatone wavelets. I can recommend!
- Elias
What software do you use use to test this way and create these nice graphics?
What software do you use use to test this way and create these nice graphics?
look at his tag line
That's one reason dipoles sound good in real rooms: the figure-8 dispersion means first reflections off the side walls are down a fair bit compared with those from the usual monopoles (horns excepted, of course).
Yes, they remain narrow into the range Toole graphs as contributing to "envelopment". Problem is that they tend to bubble in the "image shift/ source broadening" area(the areas overlap in the 500-1kHz region). Sort of the opposite of a CD type design. The Note breaks the mold in that respect. It would be most cool to hear it just once. I'd also like to hear a dipole with an extremely damped front wall.
Dan
Dan
Glanced at that write-up and came across;
which is completely incorrect, so I stopped reading.
In the "statistical region", i.e. above the Schroeder Frequency, the room statistics are independent of frequency and only have a small dependence on the reverb time, but nothing else. Getting this wrong is so common in the dogma about "room correction".
The reason that we are talking about variable time windows has to do with the ear, not the room.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.