To generate a general form of a speech sound, an impulse train or noise can be used.
In order to do that, a spectral envelope must be learnt for the specific sounds from data.
Find a speech dataset with annotated sounds.
Fit spectral envelopes to the data with the annotated sounds.
Use the spectral envelopes to generate (very simple) speech sounds.
To generate a general form of a speech sound, an impulse train or noise can be used. In order to do that, a spectral envelope must be learnt for the specific sounds from data.