Segmental quality assessment by pseudo-words

A synthetic speech segmental quality assessment method based on a non-sense pseudo-words list is proposed. The pseudo words are automatically generated by a stochastic automaton whose transition diagram represents the phonotax of the language, in terms of syllabic, stress and morphological constraints [Falaschi 87]. Full coverage of the phonemic environments and functionalities is ensured by the use of a compacting selection procedure [Falaschi 90]. Listeners phonemic confusion is automatically scored by a recognizer performances evaluation algorithm. Learning effects can be avoided by generation of new non-sense lists. Moreover, the method is naturally suited for the definition of typical complexity, phonologically constrained, test sets across different languages.