Windowing filters

Special Interest Groups Audio Processing
1

As of version 2.2.2, Audacity offers the following filters for spectrograms:

Rectangular
Bartlett
Hamming
Hanning
Blackman
Blackman-Harris
Welch
Gaussian(a=2.5)
Gaussian(a=3.5)
Gaussian(a=4.5)

The properties of those filters vary. For example, Hamming has a lower first sidelobe than Hanning, but the volume of the sidelobes decreases slower.

There’s this document that describes some more windowing filters, again, with varying properties.

What windowing filters do you like to use for your spectrograms?

2

I think the Wikipedia article on window functions is easier to read: https://en.wikipedia.org/wiki/Window_function

The Hann window is a good default choice. Other window functions may be used if there’s a definite reason why it is more appropriate.

3

That Wikipedia article contains not all the window functions described in https://apps.dtic.mil/dtic/tr/fulltext/u2/a085032.pdf. I even found a windowing function that has -42dB/octave decay, what this means is the volume drops by a factor of 128 for each octave of frequency distance from the main frequency; for reference, Hanning has a -18dB/octave decay; this doesn’t necessarily mean it’s better to use this windowing function.

In theory, the sinc function window would produce no spectral leakage. However, the sinc function is of infinite length; it looks like instead of frequency spectral leakage, this theoretical window has temporal spectral leakage (spectral leakage in time).

Choosing the window function definitely involves various tradeoffs regarding the impulse response of both time and frequency.