Hi again, Felipe,
In your last, you wrote:
Yes, one or two repetitive cycles should be enough, to compare with real speech. Though it would be useful only for the vowels, and for refinements, optimization, etc, in order to approach specific voices.
My idea is to "mimic" someone's voices, in the future. I could have, for example, Maddona singing in Esperanto? Hehehe...
In fact a comparison of the speech waveform is not very helpful in imitating a particular speaker's voice. This is because it is the spectral content, pitch pulse shape and changing frequency, and the spectral dynamics, that are the main determinants of similarity. The speech waveform can vary significantly without changing the perception of similarity because the speech waveform also depends very much on the phase relations of the harmonic components and the human ear is not really sensitive to phase.
Warm regards.
david
--------