Overlap-add
Chopping a long signal into blocks, processing each block on its own and summing the overlapping tails afterwards.
The original problem is convolution. Convolving a minute of audio with a long filter directly is expensive, and an FFT needs finite blocks. Overlap-add slices the input into blocks, zero-pads each one so circular convolution doesn't wrap around on itself, multiplies in the frequency domain and adds the results back with overlap. Its sibling, overlap-save, gets the same answer by discarding the contaminated samples instead.
The same idea underpins STFT resynthesis: the windowed frames have to sum to a constant — the COLA condition — or the level pumps in time with the frames. A Hann window at 50 % overlap is the default choice. Without overlap-add, convolution reverb and time-stretching would be a good deal slower than they are.