ABSTRACT:


	   WAVEFORM MODELS FOR DATA-DRIVEN SPEECH SYNTHESIS

			       Michael W. Macon

	  Department of Electrical and Computer Engineering
	       Center for Spoken Language Understanding
		      Oregon Graduate Institute
			      Portland, OR  USA

			      macon@ece.ogi.edu
		         cslu.cse.ogi.edu/tts


Many "data-driven" models for synthesizing speech rely on
concatenating waveforms extracted from a database.  However, the
number of perceptually-important degrees of freedom in speech make it
unlikely that enough data could be collected to cover all combinations
of phonetic variables.  By utilizing models that can transform the
waveform in perceptually-relevant ways, the space of acoustic features
covered by the data can be expanded.  The minimal requirement for such
a model is parametric control of the fundamental frequency and
duration.  In addition to this, dimensions such as voice quality
characteristics (breathiness, creak, etc.), phonetic reduction, and
voice identity can be altered to expand the range of effects
realizable from a given database.

A few classes of models have been proposed to allow varying
degrees of control over these dimensions.  Tradeoffs between
flexibility, fidelity, and computational cost exist with each.  This
paper will describe common threads running through the best-known
approaches, including handling of non-periodic components and
sensitivity to measurement errors.  Although each of these approaches
has its own merits, none have been proven to simultaneously satisfy
all the desired properties for synthesis.  We will demonstrate the
limits of some of the simplifying assumptions underlying these
methods, and outline areas needing improvement.