academic reporting of Audacity

[k.p.britton]

I am doing some academic research, and have been using audacity to record and process the sounds of some fish.I want to know how to report this in academic language, but to be honest, all the numbers are goobledegook to me.

here is an example of what is said in another academic paper:

Sounds were digitized at 441 kHz (16 bit resolution) and analysed with the Avisoft-
SASLAB Pro 433 software (1024-point Hanning window fast Fourrier transform,
FFT). Temporal features were measured from oscillograms, and frequency variables
were obtained from power spectra (filter bandwidth 300 Hz, FFT size 256 points, time
overlap 9687% overlap and a flat top window).

I am guesing that some of this info will be the same for me. I see the digitisation is at 44.1kHz for my recordings. The settings I have been using for spectrum analysis (Hanning window, 256, linear frequency)

I can see the effect of changing 256 to a different value, but can someone explain in basic terms whatthis is actually doing?

[steve]

Sounds were digitized at 44.1 kHz (16 bit resolution) - or whatever you are using
and analysed with Audacity version ??? software - the Audacity version that you are using
(1024-point Hanning window fast Fourrier transform, FFT).
Spectral data was obtained by plotting data using:
Function (window type)
FFT size (suggest you increase this to at least 1024 - preferably much higher for greater resolution
Axis (linear or log scale)

I can see the effect of changing 256 to a different value, but can someone explain in basic terms whatthis is actually doing?

In simple terms - no not really - it is complicated.
A kind of rough idea is that FFT samples the audio into a number of buckets according to frequency. The different window functions are like different size/shape/colour buckets. The greater the FFT size, the more buckets you have, so the finer the resolution.

[k.p.britton]

the FFT max size is presumably set by the length of the sample your analysisng, and as some of the things I am looking at are very short (less than 10 ms) I dont think I have the option of this?

[steve]

Yes it is, but you may be able to repeat the sample to make it longer.
To repeat a selection, select "Repeat" from the Effects menu.

If you want to boggle yourself with the science, there is some information about Plot Spectrum in Audacity 1.3.8 (which is slightly different from previous versions) discussed here: http://n2.nabble.com/Re-Question-about- ... l#a3192212

The PDF file that is referred to requires a recent version of Acrobat Reader with an additional plug-in to be able to open it. It contains detailed explanations that fly straight over my head.

[k.p.britton]

I managed to get the pdf open, but it was also wellover my head.

I have a question about processing, which follows from this. I am recording fish sounds in a fairly noisy (and VERY large) aquarium. I want to see the fish sound spectrum without the background. If I take the fish sound spectrum, and export it, so I have a value for each bin, then take a similar length sample from before the fish sound occured to represent "background", and subtract the individual bin values of the second sample from the first. do I have a meaningful "residual" value for frequebncy of the fish sound?
I understand that 3dB variance reprsents a doubling of acoustic power. (this is presumably why 3dB below the max power peak is used todefine the bandwidth of the signal and this Q3dB
tuning value).
The values I am getting from the "total"spectum vary from -50 to around -20 dB, with the background between -50 and -30/40 so I get a spike inn difference of around 15 to 20 over a reasonably narrow frequency band (say 600-800 Hz, using the -3dB cuttoff).

i dont particularl intent to publish the data from these calcualtions apart from using it to generate frequency distributions, bandwidths and Q3dB, perhaps.

[steve]

I managed to get the pdf open, but it was also wellover my head.

Like a jet plane at 40 thousand feet - yes, I know what you mean

If I take the fish sound spectrum, and export it, so I have a value for each bin, then take a similar length sample from before the fish sound occured to represent "background", and subtract the individual bin values of the second sample from the first. do I have a meaningful "residual" value for frequebncy of the fish sound?

That sounds reasonable, but really I don't know. What you are doing there is a bit like the "Noise Removal" effect. It may be interesting to try the Noise Removal effect (use the background noise as the noise profile and apply it to the recording of the fish chatting) and then compare the analysis from all three.

It may be worth you writing to the Audacity Users Mailing List about this - some of the clever people that understand about FFT sometimes drop in there, so you may get lucky. http://audacityteam.org/contact/lists

[k.p.britton]

i tried the noise removal, but its just TOO aggressive!!

anyway, I went to see another academic who works with sound today 9i was going to borrow a higher quality hydrophone from him) andhe suggested sidestepping the whole issue. he thinks the data i have is very good, but the frequency spectra are shite (not his words).. ans so I just need to skirt around it as i write up. sounds good to me!!

[steve]

i tried the noise removal, but its just TOO aggressive!!

Even when turned down to minimum?

If so, there is a little trick to make it more gentle - before you make the Noise Profile, reduce the level of the audio that you are using as the "noise" sample (use the Amplify effect with a negative amount of amplification).

BTW, it looks like you are not the only one interested in FFT this week: http://forum.audacityteam.org/viewtopic ... 28&t=12836