Working paper Open Access
Beckman, Mary E.; Hertz, Susan R.; Fujimura, Osamu
SRS (Speech Research System) [Hertz 1982] has been used recently to write a set of phoneme-based rules for Japanese. This rule set tests several hypotheses about Japanese phonology. In particular, it tests the usefulness of a hierarchical phrase-structure analysis for generating pitch patterns. The user demarcates the phrases in the romanized input string with boundary symbols, which have associated boundary features.
These boundary features, together with user-provided accent marks, are used to generate a segment-by-segment specification of pitch features, such as [low] (low-pitched) and [acc] (accented).
Synthesizer parameter rules then refer to the features of the boundaries and of the intervening segments to produce various parts of an utterance's pitch pattern, such as phrase-final lowering or the post-accentual fall. Our SRS pitch rules for Japanese have produced natural-sounding intonational and accentual patterns in a large variety of utterances.