Assigning Intonational Features in Synthesized Spoken Directions

Assigning Intonational Features in Synthesized Spoken Directions ° James Raymond Davis The Media Laboratory MIT E15-325 Cambridge MA 02139 Julia Hirschberg AT&T Bell Laboratories 2D-450 600 Mountain Avenue Murray Hill N3 07974 Abstract Speakers convey much of the information hearers use to interpret discourse by varying prosodic features such as PHRASING, PITCH ACCENT placement, TUNE, and PITCH P.ANGE. The ability to emulate such variation is crucial to effective (synthetic) speech generation. While text-to- speech synthesis must rely primarily upon structural in- formation to determine appropriate intonational features, speech synthesized from an abstract representation of the message to be conveyed may employ much richer sources. The implementation of an intonation assignment compo- nent for Direction Assistance, a program which generates spoken directions, provides a first approximation of how recent models of discourse structure can be used to control intonational variation in ways that build upon recent re- search in intonational meaning. The implementation fur- ther suggests ways in which these discourse models might he augmented to permit the assignment of appropriate intonational features. Introduction DIRECTION ASSISTANCE ! was written to provide spo- ken directions for driving between any two points in the Boston areal7] over the telephone. Callers specify their origin and destination via touch-tone input. The program finds a route and synthesizes a spoken description of that route. Earlier versions of Direction Assistance exhibited notable deficiencies in prosody when a simple text-to- speech system was used to produce such descriptions[6], because prosody depends in part on discourse-level phe- nomena such as topic structure and information status which are not generally inferrable from text, and thus *The inton~tion,d component described here was completed at AT&T Bell Laboratories in the summcT of 1987. We th~nk Janet Pie~Tehtunbert and Gregory Ward for valuable discussions. 1 Direction Assistance was originally developed by Jim Davis and Tom Trobaugh in 1985 at the Thinking Maf_~ines Corporation of Cambridge. cannot be correctly produced by the text to speech sys- tem. To alleviate some of these problems, we modified Direc- tion Assistance to make both attentional and intentional information about the route description available for the assignment of intonational features. With this informa- tion, we generate spoken directions using the Bell Labo~ ratories Text-to-Speech System[21] in which pitch range, accent placement, phrasing, and tune can be varied to communicate attentional and intentional structure. The implementation of this intonation assignment component provides a first approximation of how recent models of discourse structure can be used to control intonational variation in ways that build upon recent research in into- national meaning. Additionally, it suggests ways in which these discourse models must be enhanced in order to per- mlt the assignment of appropriate intonational features. In this paper, we first discuss some previous attempts to synthesize speech from representations other than sim- ple text. We next discuss the work on discourse structure, on English phonology, and on intonational meaning which we assume for this study. We then give a brief overview of Direction Assistance. Next we describe how Direction Assistance represents …

Content truncated to 500 words