Posts written by angelo.farina

Today (4 January 2023) a new version of the STI plugin for Adobe Audition has been released, named STI_2020.xfm
The new plugin does not overwrite the previous one, named STI.xfm, which was operating according to the previous version of the standard (2011):
please not that, for the measurement to be compliant with the new standard, the spectrum of the artificial mouth, measured at 1.0 m in front of it, must now be equalised to the following target spectrum:


Then the measurement is performed as usual, with the indirect method, as explained here.
At the end the results are shown according to the new standard, which now only provides a single value for STI (male) and of STIpa, as shown here:


The new STI module version 4.6 can be downloaded (together with an updated release of the old version 4.5) form the following folder, which also contain the latest version 4.5 of several other Aurora plugins (all these are recommended over the previous version 4.4):
http://pcfarina.eng.unipr.it/Aurora/download/Aurora45-Alpha/

It appears that the Zylia drivers are 64-bts only, and Audition 2.4.1 is still 32 bits.
SO the solution is to use Audition 3.xx for making the synchronous Playback and recording, and then switch back to Auidition 2.4.1 for doing the processing.
If you like Reaper, you can use Reaper for doing everything.
You can use two instances of X-MCFX plugin, the first for converting the 19-chanenls to 16-channels Ambix, the second for performing the sweep deconvolution.
Here is X-MCFX: www.angelofarina.it/X-MCFX.htm

Audacity 2.4.1 Mac only works up to Catalina. For newer versions of Mac OS, I recommend using Playonmac and running the Windows version of Audacity 2.4.1. Works great on my Mac M1 Max running Monterey...
Even better is, also on PlayonMac, running the good old Audition 1.5 with the XFM version of the Aurora plugins. This is definitely my favourite setup...


Aurora has not yet be recompiled for Audacity 3.xx. Audacity 2.41 is 32 bits on Windows, 64-bits on Mac.

Sorry, Aurora does not support the obsolete interrupted noise method for measuring the reverberation times.
The ISO3382 Acoustical Parameters module is designed for processing impulse responses.
Either measured with an impulsive sound source (pistol, balloon, firecracker, etc.), or, better, using a loudspeaker driven with the Exponential Sine Sweep signal.
Aurora contains the tools required for generating the ESS test signal, the matched Inverse Sweep, and for performing the convolution of the recorded signals with the Inverse Sweep, obtaining the impulse response.

First of all, there is no need to use Reaper at all.
Reaper is complex to use, and it is very easy to mess up everything!
Audacity can be used for playing back the sweep and simultaneously recording your 19 channels from the Zylia microphone array. Just remember to mute the track containing InvSweep while playing and recording...
After doing the recording (which must last some seconds after the end of the sweep), select all, then invoke the Aurora Convolution plugin.
Place the recorded signals in Audio Data, place the Inverse Sweep in Filters, and perform the convolution.
You will get a set of 19 new tracks, each of them containing one channel of the impulse response, with a lot of silence before. If you want, you can select and cut away (or delete) the leading silence, for exactly the same length as the sweep (say, 10 seconds). I usually also cut away the silence AFTER the IR has faded to zero.
Now it is time to convert from A-format (19 channels) to B-format (16-channels, Ambix 3rd order).
You need the 19x16 FIR filter matrix which you can download from my web site, here:
http://www.angelofarina.it/Public/Xvolver/...Zylia-Jul-2020/
If your microphone is the new model, you need to downloa0d the WAV file named A2B-Zylia-3E-Jul2020.wav
After importing also this in Audacity, you again select all and invoke the Aurora Convolver plugin.
You drag the 19 impulse response channels to Audio Data, the 16 channels of the FIR filter matrix to Filters, select the flag for "matrix mode" convolution and perform the matrix convolution.
You will get 16 new channels, each containing one of the Ambix channels.
You select these 16 tracks and export them in a single WAV multichannel file. Of course, you must enable the multichannel export capability of Audacity (by default it downmixes everything to 2-channels stereo).

The latest version of Aurora is only 1/1 octave bands. However we have some old versions with were working in 1/3 octave bands, and which you can download from the Aurora web site. We abandoned this, because the usage of 1/3 octave bands requires much longer averaging times, and in general produces larger errors.
Better to have a smaller number of spectral bands with accurate results, than many more spectral bands containing values with larger uncertainty...

I got access to the 4files and checked if they works.
I could not complete the STI calculation, as you did not post the IR, just a sine sweep, which cannot be processed without its own matched Inverse Filter signal..
I assumed the calibration was done with a 94 dB standard calibrator (the calibration WAV file name usually always contains the explicit indication of the calibration level in dB !!!)
After calibrating at 94 dB (the Full Scale value resulted 111 dB) I did the SPL analysis of your WAV files.
The levels, indeed, are low, very low, so something seems to be wrong in your measurement setup. Perhaps you did change sone gain between the calibration and the recording...
Of course you should not change anything after recording the calibration signal!
But the results are NOT negative dB. so you Windows system is badly misconfigured.
I looked in the images you sent me, and possibly I did find the problem: you appears to be using an horrible and unstable FRENCH version of Audacity, and possibly also using a terrible FRENCH version of Windows, with FRENCH international parameters (comma instead of decimal point, etc.)....
EVRUTHING UST BE IN ENGLISH for scientific programs to work reliably !!!
As a starting point, set the International Parameters of Windows to English-UK mode, so that the decimal dot is used, and try again.
But I warmly recommend that everyone always employs always and only native English version of the operating system and of any program employed.

The point i that your explanation was impossible to understand.
IACC should measured after recording a room impulse response with a binaural microphone...

In the meanwhile, I did the STI processing using "my" set of 4 WAV files, which I linked in my previous message.
First I imported them in Audacity:

As you see, I selected all 4 of them (CONTROL-A).
Then I invoked Tools - Aurora STI and selected the Fullscale Calibration label.
Here I choose which of the 4 files contains the calibration recording, I specify the calibration level in dB (68 dB in this case, but usually it is 94) then I press the button "Calibrate" and here is what I get:

Now I go to the next label, titled SNR Calculation. Here again I specify the file names for the background noise and for the Signal+Noise WAV files:

And I click on the "Compute Levels" button. A popup window appears, showing the computed spectra of Background Noise, Signal and Signal+Noise:

As you see the levels are NOT zero... And after closing this popup window, the new SPL value for Background Noise, Signal and Signal+Noise levels are shown in the table.
Finally I go on the Room Impulse Response label, and again I select the file name containing the IR:

Now the plugin has all the data required for computing STI. So I click on the "Compute STI" button, and the final window showing the STI calculation results is shown:


If you give me access to your 4 WAV files, I can repeat the same processing with them, and see what is wrong...

The Google Drive folder is not accessible.. I did "request access" to it...

A means the A-weighting filter applied to the signal.
L means linear, no filter. On modern sound level meters it is called Z.
Both are "wide band" values.
The octave-bands values are obtained applying 6-poles pass-band filters, compliant to IEC standard for octave-band filters.
You can get the filtered version of the impulse response being processed by selecting one of the frequency bands (or A, or L) and then pressing the button "filter" on the low-right instead of "close".

Edited by angelo.farina - 9/11/2021, 18:52

On Mac OS, better to use the good old Audition 1.5 (which works perfectly thanks to Playonmac) and the XFM version of Aurora plugins...
The version for Audacity was never properly debugged.
However, Tau IACC is the time where the maximum occurs (it should be 0.0ms, but it can be a value in the range +/- 1.0 ms if the dummy head was not pointing exactly to the sound source).
For an explanation of the meaning of Tau IACC and wIACC, please read this paper by Dario D'orazio at page 4 and fig. 5:
http://acustica.ing.unibo.it/Staff/paolo/p.../23-aes2009.pdf



Edited by angelo.farina - 9/11/2021, 15:19

Yes, this is possible, but never expect negative values, as I have explained the idea is to find the maximum value of the CC function on the interval between -1ms an +1ms. Even if you have a strong negative peak, this will not be recognised as the IACC value, as this is defined as the maximum value, and the negative peak is a minimum!
The best thing is to look directly at the CC function, and not relaying on the automatised extraction of the IACC value.
The CC function is trivial to obtain, just convolve the first 80ms of the impulse response measured at the left ear with the first 80 ms of the impulse response measured at the right ear, time reversed.
Example: this is a binaural impulse response:

Here I trimmed just the first 80ms:

Now I select just the Left channel, and copy it into a new WAV file:

It is time now to compute the Autocorrelation function, so I copy this Left IR on the Windows clipbpoard, and I invoke the Convolve with Clipboard Aurora plugin, selecting to time-reverse the IR, and to autorange the results so that the maximum is 1.0:

The result is being rescaled by a proper amount, namely -39.46 dB:

And here we see the AutoCorrelation function of the Left channel:

As expected, the maximum peak is at 1.0 and at 0.08 s (that is "zero time" for the correlation functions).
WE now "undo" the convolution, coming back to the left IR, and we copy onto the Windows clipboard the Right channel IR, selecting just it in the stereo IR:

And copying it onto the Windows Clipboard.
Now we are back to the Left-Ear IR:

And I convolve it with the Windows clipboard, which now contains the right-ear IR. Again, using the time-reversal of the IR, but de-selecting the auto-gain, se we employ the fixed gain of -39.46 dB which was found when computing the autocorrelation:

The result is the Cross-Correlation function:

For evaluating the IACC we must zoom in the range of +/- 1 ms around the "zero time", so in this case the time range is 0.079 to 0.081 s:

As you can see, in this case the max value is approximately 0.30, and the time of this max is substantially 0.0 ms. The sound source was in front of the dummy head...

For attaching a screenshot you need to upload the image on an external server, I suggest postimages: https://postimages.org/
Then you paste here the code containing the link.
I still do not understand what you are attempting to do.
IACC is an acoustical parameter defined in the ISO 3382 standard, to be measured inside a concert hall. There is only ONE loudspeaker when measuring IACC (it should be omnidirectional, dodecahedron, etc.).
The loudspeaker is on the stage, and the dummy head is in the audience, pointing EXACTY towards the sound source.
Only when this is done properly you get meaningful values of the IACC parameter, in frequency bands of 1 kHz and above...
IACC cannot be negative, as it is defined as the maximum value of the Cross-Correlation function evaluated in the range of -1 to +1 ms, hence as the curve is fluctuating between negative and positive values, and the maximum is taken, it is always positive.
What happens when the sound is not coming exactly from the front direction is that the time where this maximum occurs is not at 0 ms, but it is slightly left or right of the zero point on the time axis.

Yes, you can attach a screenshot. But I do not really understand what you are doing...
So please, explain better what you are attempting to measure, and describe step by step your procedure.