When reading the methods sections of bioacoustics papers, I often see a window length and % overlap reported in the description of how the spectral analysis was done.

  • What is the purpose of this?
  • What factors inform what an appropriate % of overlap between neighbouring windows should be?
  • Does it have anything to do with addressing 'spectral leakage'?

Thanks in advance


2 Answers 2


This answer assumes language of FFT based processing (e.g. Spectrograms). For other type of processing language could be changed.

Assume you have a signal just between two FFT blocks. As the FFT assumes circular data, the signal constitutes a discontinuity that corrupts the spectrum. Yes this is spectral leakage, which is attenuated by a windowing function that reduces the influence of edge effects from spectrogram.

Windowing reduces in the assumed case the signal close to the edge. Consequently, if the next FFT is shifted by a complete window, the signal may be missed, but when shifted by 1/2 the window length, the signal will fully captured.

Overlapping windows by 50 % is standard for Hann, Hamming, and other standard window functions.

More overlap is useful when one wanted a smoother (higher correlated) spectrogram.

Note: to estimate power spectral density (PSD), window shape and overlap must be considered.


Temporal resolution in a spectrogram is fundamentally determined by the window size, equal to the FFT-length. In a simple plot of the spectrogram of a short signal, one pixel on the screen equals one window. If a high frequency resolution is wanted, which requires long windows, the spectrogram appears pixelated on the time axis. This can be smoothed by making windows overlap. 50% overlap gives twice the display resolution, 90% overlap 10 times the display resolution, which creates visually more appealing displays. When interpreting the spectrogram, this "smoothing" must be kept in mind, as the fundamental time resolution is still given by the FFT-size/window length and cannot be improved by overlapping. Windows are correlated, as explained in @WMXZ's answer.

If a Hann window is used, one can argue that 50% overlap is well justified, because no part of the signal contributes more to the spectrogram than others. In the individual Hann-window, the middle part is weighted with 100%, but the beginning and end are weighted 0%. However, the sum of a series of 50% overlapping Hann-windows equals 1 except for the first half and last half window, meaning that every part of the signal contributes equally. If larger overlap is used, the overlapping segments contribute with more than 100%, and less than 100% for smaller overlaps. Same argument can be used for overlapping windows in power spectrum estimates.

This symmetry a particularly useful feature of Hann windows, generally not true for other windows (Bartlett/triangular window is one exception), making Hann-windows a very good first choice for analysis.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.