Word stress assignment in a text-to-speech synthesis system for British English

Abstract

A technique for the assignment of word stress to isolated British English words is described. The input form of the word contains both morphological and phonological information. The input phonological form is more abstract than the normal phonemic transcription. All vowels are shown in unreduced form, allowing the stress rules to condition the subsequent reduction. The morphological analysis is based on a simple set of morpheme types. The input string is parsed into stress foot units. The stress foot units form the basis for the assignment of primary and secondary word stress. Unstressed vowels are reduced and certain other vowel modifications are performed. The results of testing the technique are compared with the performance of an affix-based method. The technique accounts for slightly more data than this taxonomic approach; this is because the latter uses a morphological shorthand to encode what are essentially phonological facts. The performance of the technique is also compared with that of two further methods and is found to be more satisfactory on both the theoretical and the practical levels.