noverlap argument to spectrogram

The documentation for noverlap says that it indicates the "number of overlap samples." But this seems unclear. Suppose I have an noverlap of 1. Does that mean that a given window will overlap exactly one sample with the following window and 0 with the preceding, or that it will overlap one sample with the preceding and succeeding windows? If it is the former, that seems odd, because it would mean that for odd noverlap, different windows would look different with regards to overlap (you would have pairs of windows that overlap, but no pair overlaps with another. This also creates an ambiguity if you have an odd number of windows), but if it's the latter then the documentation is technically incorrect, so which is it?

Answer 1

[Daniel Shub]

While I haven't tested it, and I cannot think of an easy signal that would allow me to, I believe that noverlap refers to the amount of overlap of window N with window N-1.

[Steven]

I came up with some tests, and determined that you are right:

>> S=spectrogram([1:10000], 20, 0); size(S)

ans =

129 500

>> S=spectrogram([1:10000], 20, 5); size(S)

ans =

129 666

>> S=spectrogram([1:10000], 20, 10); size(S)

ans =

129 999