Voice Modification using the Sinusoidal Analysis/Synthesis System
Senior Scientist, Nellymoser Inc.
An implementation of the algorithm for pitch scaling by re-sampling and time scaling is developed using the sinusoidal analysis/synthesis system. The usual interpolation problems are completely avoided such that the pitch-scale factor and the time-scale factor can be arbitrarily time varying. The pitch-scaled, time-scaled speech is of high quality and provides a basis for performing experiments in voice modification. Although the resulting algorithms are conceptually quite simple to implement, to do so has required utilization of the sine-wave based pitch extractor, voicing detector, phase model, harmonic vocoder model, cubic spline and allpole spectral fitting, as well as the usual analysis and synthesis structures. These algorithms will be reviewed to provide a comprehensive basis for the time scaling, pitch scaling and vocal tract transformations required for a voice modification system using the sinusoidal speech model.