Windfarms & ETSU-R-97 tonality checking

[fcluson]

I am using the Audacity to examine for tonality of wind farm noises

The standard method is described in ETSU-R-97 - http://www.hayesmckenzie.co.uk/download ... hable).pdf

look around pages 100

Is there a feature - or can I ask for one please - to allow the splitting of the frequency into 2Hz buckets (page 105 of the above)

[kozikowski]

Not Found

The requested URL /downloads/ETSU Full copy (Searchable was not found on this server.

Additionally, a 404 Not Found error was encountered while trying to use an ErrorDocument to handle the request.

Apache Server at http://www.hayesmckenzie.co.uk Port 80


Koz

[fcluson]

Sorry about that - lets try again - I think it got truncated

http://www.hayesmckenzie.co.uk/download ... hable).pdf

[Robert J. H.]

I am using the Audacity to examine for tonality of wind farm noises

The standard method is described in ETSU-R-97 - http://www.hayesmckenzie.co.uk/download ... hable).pdf

look around pages 100

Is there a feature - or can I ask for one please - to allow the splitting of the frequency into 2Hz buckets (page 105 of the above)

I'm not sure if I understand this procedure correctly.
It seems that a 2 minute recording is being analysed.
The sampling rate should be presumably 4000 Hz (2 kHz bandwidth)
The signal is split into FFT frames of 0.29 to 0.4 s length with proper windowing applied (do they overlap or not?).
The averaged Rms spectrum is searched for prominent peaks.
Around this peaks, the critical band is defined according to the formula

Code: Select all

(+ 25 (* 75 (expt (1+ (* 1.4 (expt (/ freq  1000.0) 2.0))) 0.69)))

or simpler as 100 Hz below 500 Hz and 20% of the center frequency above 500 Hz.
which means that a certain amount of FFT bins will be recursively analysed.
This leads to the conclusion that your "buckets" are not fixed.
This first step could be done in Excel or similar.
You just have to export the 2-min spectrum (via Analyze, Plot Spectrum).
However, the procedure is complicated by the fact that the detected tones are refined by taking the average of 10 % of the individual spectra.
Thus, you (or we) have to write a Nyquist plug-in to make the whole analysis.
Not very easy, I'm afraid.

[fcluson]

Thanks for that

lets focus on the sub 500Hz range - I am new to Audacity but have programmed many other apps elsewhere

Where do I enter the code you gave me ?

[Robert J. H.]

Thanks for that

lets focus on the sub 500Hz range - I am new to Audacity but have programmed many other apps elsewhere

Where do I enter the code you gave me ?

Effect menu --> Nyquist Prompt
It is basically a LISP interpreter with extended sound support. It is intended to test some code on the fly (although a whole 1000 line program could be run from it).

Let's apply the formula to all frequencies from 100 to 2000:

Code: Select all

(dotimes (i 20)
   (setf freq (* 100 (1+ i)))
   (format t "Center Frequency: ~a, Band Width: ~a~%" freq (+ 25 (* 75 (expt (1+ (* 1.4 (expt (/ freq  1000.0) 2.0))) 0.69)))))

You have to press "Debug" to see the results.
However, the joined Nordic procedure seems to simplify this to (max 100 (center * 0.2)).

Julius O. Smith (*) gives yet another approximation.
Here are all at once:

Code: Select all

(dotimes (i 20 "See debug screen for results.")
   (setf freq (* 100 (1+ i)))
   (format t "Center Frequency: ~a~%Band Width ~a JOS: ~a, Nordic: ~a~%" freq 
     (+ 25 (* 75 (expt (1+ (* 1.4 (expt (/ freq  1000.0) 2.0))) 0.69)))
     (+ 94 (* 71 (expt (/ freq 1000.0) (/ 3 2.0))))
     (max 100 (* freq 0.2))))

Those values have to be rounded, according to the FFT resolution.
Recommended is a bin-distance of 2.5 to 3.5 Hz.

(*) Julius O. Smith:

Fall 2004 to the present: Professor of Music and Associate Professor (by courtesy) of Electrical Engineering, Stanford University, based at the Center for Computer Research in Music and Acoustics (CCRMA). Activities include teaching courses in signal processing and music technology, graduate student advising, and research in signal processing techniques applied to music and audio.

[fcluson]

Jules

thanks - let me have a go and I'll come back to you - really appreciate the help

[kozikowski]

Do you have a known event to analyze? Do you have a known test condition and the expected result?

Koz

[fcluson]

Koz

I have a number of sound files ranging from 10 to 30 mins . I have done a Spectrum lg and its clear there is a line at around the 168Hz range. I have done a FFT with different "bins" and there again is a clear spike. I exported the FFT to excel which is when I found the buckets where not evenly spread.

So yes - I know there is an event to look for and I have seen evidence of it in tools which are already available. I now need to be more precise.

Expected results - no - but I am sure that there is a tone as per the ETSU document I just need to quantify it as I want to determine the penalty.

I will want to lay the results in fron of the wind farm owners and the local council - and they will have expensive professionals looking at my results so I want them to stand up scrutiny

[fcluson]

Koz - would you like me to post a snipit of the data so that someone can have a look at it ?