Overlap-Save Method

The classical overlap-save method [187], unlike OLA, uses no zero padding to prevent time aliasing. Instead, it

discards output samples corrupted by time aliasing each frame, and
overlaps the input frames by the same amount.

More specifically:

  • If the input frame size is $ N$ and the filter length is $ L\ll N$, then a length $ N$ FFT and IFFT are used.
  • As a result, $ L-1$ samples of the output are invalid due to time aliasing.
  • The overlap-save method writes out the good $ R=N-L+1$ samples and uses a hop size of $ R$, thus recomputing the time-aliased output samples in the previous frame.

The name ``overlap-save'' comes from the fact that $ L-1$ samples of the previous frame are effectively ``saved'' for computing the next frame.

Next Section:
Time Varying OLA Modifications
Previous Section:
Dual of Constant Overlap-Add