Please pardon me. Maybe I didn't make myself understood.
If explainiing STFT as linear filtering, then its hop size
must be set according to the type of analysis window and
widow length so that the original signal can be restored from the time
downsampled STFT. It is exactly right!
My question is:
The outputs of phase vocoder are time-variable ampltitude
(or say, instantaneous amplitude) and instantaneous frequency.
In the discrete-time case, instantaneous frequency have to be
computed as difference of phase or as given by
(J. L. Flanagan and R. M. Golden, ¡°Phase vocoder,¡±).
In the first case, phase must be extended beyond
the principal value interval [-pi,pi]. If the hop size
is 1, this job is easy. If not, how to do then?
In the latter case, differential of STFT's real part and
imaginary part have to be estimated by discrete-time
filtering of them, for example, the first order for ward or backward
difference. If the hop size is not 1, differential can still be replaced
by difference? The real part and imaginary part of STFT are
band-limited in what extent? How to set a reasonable hop size?
Superk (Kun She)