Sixth International Conference on Spoken Language Processing
October 16-20, 2000
A Unified View on Synchronized Overlap-Add Methods for Prosodic Modifications of Speech
Werner Verhelst, Dirk van Compernolle, Patrick Wambacq
ESAT/PSI, Katholieke Universiteit Leuven, Belgium
We present several synchronized overlap-add methods for
prosodic modification of speech in a unifying framework.
The discussion starts by exploring time-scale modification
based on the short-time Fourier transform and proceeds to
review an iterative phase reconstruction method for time-scaled
magnitude spectrograms. The search for a good initial
phase estiniate leads to waveform.synchronized overlap
add methods for time-scaling of speech, in particular SOLA
and WSOLA. Pitch modification is made possible when
pitch-synchronized overlap-addition is applied to wideband
short-time Fourier transforms, as is the case with current
techniques like PSOLA, PIOLA, and MBROLA. All these
prosodic modification methods essentially perform some
form of automatic editing of the speech waveform in the
time domain. Therefore, they can be computationally
efficient and at the same time maintain a natural sound quality.
Verhelst, Werner / Compernolle, Dirk van / Wambacq, Patrick (2000):
"A unified view on synchronized overlap-add methods for prosodic modifications of speech",
In ICSLP-2000, vol.2, 63-66.