Let's say that we want to manipulate a signal in the frequency domain using a short time Fourier transform and once we're done , we use an inverse STFT (plus overlap and save) to convert the results back to the time domain (this is basically a simple phase vocoder implementation).
Now, the problem is that this process doesn't retain phase information , which creates phase incoherence issues (this is evident in audible digital artifacts and in some cases the manipulated/reconstructed signal is clipped and completely unusable).
In my case, the STFT analysis is done using overlapping frames, which were windowed (e.g Hann, Hamming etc) prior to the FFT. This alone (the overlapping and windowing) creates phase incoherences, and the problem is further compounded by manipulating spectral content.
So, are there any standard ways/strategies to deal with this kind of situation?
Thank you
No comments:
Post a Comment