| Value | Meaning |
|---|---|
| right | The window is asymmetric and if it goes to zero (eg: HANN), it will have one zero at the FIRST coefficient output[N-1]. Its center is exactly at output[(N/2)-1] which is an integer delay for even window lengths. This loose one sample of latency, however it has the easiest to compute latency in most usage. |
| left | The window is asymmetric and if it goes to zero (eg: HANN), it will have one zero at the LAST coefficient output[N-1]. Its center is exactly at output[(N/2)-1] which is an integer delay for even window lengths. This has the best latency characteristics, however this might make the latency computation itself trickier. |
| symmetric | The window is symmetric and if it goes to zero (eg: HANN), it will have two zero coefficients. Its center is exactly between samples at position output[(N - 1)/2] which is NOT an integer delay for even window lengths. Such a window might also break derivativeness. However FOR HISTORICAL REASONS THIS WAS THE DEFAULT SETTINGS. IT IS ADVISED NOT TO USE IT, FOR THE CONSEQUENCES ARE SUBTLE AND OFTEN HARMFUL. |
How "aligned" the window is in its support. Very important, see Issue #236. When choosing a window, you probably want to take a hard look about WindowSymmetry because it has implications for latency and correctness.