A command to copy a part of each selected Sound object to the Objects window.

Settings

Time range (s)

the time range (t₁, t₂) of the part that has to be extracted.

Window shape (standard: “rectangular”)

one of 12 possible window shapes; see below.

Relative width (standard: 1.0)

the relative width of the extracted part, as compared to the duration t₂ − t₁. If the relative width is 1.0, then the part from t₁ to t₂ will be extracted. If the relative width is 2.0, then the part from t₁ − 0.5 dur to t₂ + 0.5 dur will be extracted, where dur is t₂ − t₁. See below for when and why this is useful.

Preserve times (standard: off)

if on, then the time domain of the resulting extracted Sound object will run from t₁ to t₂. If off, then the resulting extracted Sound will be time-shifted to start at zero, i.e. its time domain will run from 0 seconds to t₂ − t₁.

Window shapes

Suppose we have the following sound, with the part we want to extract (from 0.06 to 0.16 seconds) already marked:

We now extract the part form 0.05 to 0.15 seconds:

With Preserve times off, we instead get:

These were extracted with a rectangular window. We have many other shapes (see Sound: Multiply by window...). With a Hanning window, we get:

With a Hamming window, we get:

With a triangular window, we get:

With a parabolic window, we get:

With a Gaussian1 window, we get:

With a Kaiser1 window, we get:

These last six windows have a comparable effective duration. The names of the last two windows, however, sound as if they want to approximate a Gaussian shape, but they obviously don’t (they don’t go to zero fast enough at the edges). We can improve the Gaussianness of the window shape by halving the effective duration. With a Gaussian2 window, we get:

and the very similar Kaiser2 window:

The advantage of these two windows is that they cut away too muich of the signal at the edges. To make the effective window duration similar again, we would like to extract a part that is physically twice as long. We can do that by setting Relative width to 2.0 instead of 1.0. The Gaussian2 window then results in:

and the very similar Kaiser2 window:

The Gaussian3, Gaussian4 and Gaussian5 windows approximate the Gaussian shape even better than Gaussian2 and Kaiser2 do, but they require even longer relative widths, namely 3.0, 4.0 and 5.0 respectively.

In Praat, most acoustical analyses are done with Kaiser2 windows, such as Sound: To Spectrogram... and Sound: To Pitch (ac)... (the latter with the Very accurate setting switched on).

Links to this page

• Extract selected sound (windowed)...