5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

A Nonlinear Unit Selection Strategy for Concatenative Speech Synthesis Based on Syllable Level Features

Martin Holzapfel (1), Nick Campbell (2)

(1) SIEMENS AG, Germany
(2) ATR ITL, Japan

This paper describes an improved algorithm, motivated by fuzzy logic theory, for the selection of speech segments for concatenative synthesis from a huge database. Triphone HMM clustering is employed as an adaptive measure for articulatory similarity within a given database. Stress level contours are evaluated in the context of their surrounding vocalic peaks. The algorithm uses a beam search technique to optimise the suitability of each candidate unit to realise the desired target as well as continuity in concatenation.

Full Paper

Bibliographic reference.  Holzapfel, Martin / Campbell, Nick (1998): "A nonlinear unit selection strategy for concatenative speech synthesis based on syllable level features", In ICSLP-1998, paper 0521.