naturerecordists
[Top] [All Lists]

Re: Determining peak frequencies

Subject: Re: Determining peak frequencies
From: "Avocet" madl74
Date: Wed Nov 21, 2012 5:22 pm ((PST))
> Playing with it, if I was trying to use it to determine either the
> highest or the loudest frequency in a call, I'd be wondering what I
> should set the "size" parameter to. Its value has a dramatic effect
> on the resulting graph.

Peter,

The "size" paramater is the number of samles it ues in the
calculation. The best way to see how this works in the Power Spectrum
and the Spectrogram is to try the extremes of values. With a "high
definition" setting, you need a long enough sample for the Power
Spectrum, but it tells you this.

> Obviously I don't know what that frequency is yet, so how do I know
> what to set it to? Similar problem for the highest frequency.

Essentially both the routines do the same thing, and the apectrogram
vertical plot is a power spectrum from a number of samples but is
repeated continuously.

With the Spectogram, the "sample" tradeoff is between frequency detail
and the time detail, blurring it vertically or horizontaly. Half way
does for many sounds. You don't get a precise frequency reading with
the spectrogram, but you can get a spot frequency by selecting a short
section and using the Power Spectrum which gives you an average over
that short section.

It is worth Normalising a quiet recording, or at least raising the
level for reading the data. The Spectrogram gives you a power level
indication by the colour as well as frequency and the time line, which
makes it a useful tool and shows the varying powers of each frequency
component changing with time.

As an example with a bird going "wheep", you will see a rising pale
region on the Spectrogram. The loudest paler part could be halfway,
fading towards the high frequency end. Other frequencies like hiss and
aircraft noise are recogisable as well, and you can see the effect of
a bass cut if you apply one.

> Playing with these spectrogram settings has revealed detail I never 
> realised existed in the low frequencies, and I can scan through and 
> spot birds calling softly that I hadn't noticed before.

If I want to examine a quiet call, I make an Equalisation filter,
rolling off the frequencies above and below the call on the
spectrogram and apply that to a selected section. I can then adjust
the level to give a better analysis of the wanted call.  CTRL-Z
restores the original.

One effect you can see on a spectrogram is if your mic is picking up
an echo. This produces horizontal bars across the display and a comb
effect on the power spectrum. Try mixing two mics at a different
distance to recognise this "phasing" effect and also to hear what it
sounds like. I've got a very strange echo which only shows up like
this with a prop aircraft at a particular height and direction.

> And for the highest frequency, how does one distinguish the call
> from mic hiss, etc?

Techie note. Background hiss is often an even colour on the
spectrogram, but mic hiss will normally have a characteristic tip-up
curve in the Power Spectrum in the highest KHz frequencies. This is
"thermal noise" (aka "root frequency noise") which can be from a
powered mic or preamps, or if you are lucky, actual thermal noise from
sir molecules rattling about. This is the value used in "Excess noise"
parameters and you can't do any better, full stop. Feel happy if you
see this tip-up.

David

David Brinicombe
North Devon, UK
Cogito cogito ergo cogito sum - Ambrose Bierce







<Prev in Thread] Current Thread [Next in Thread>
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU