Generating Natural Speech from Text

This paper outlines progress towards speech synthesis systems which are capable of generating natural speech from ordinary text. Its theme is that although the engineering and phonetic tools for this task are nearing completion, many difficult problems of high-level linguistic analysis exist which have hardly been considered - for example, the generation of appropriate intonation information from text. These problems are highlighted by describing an existing synthesis system, paying particular attention to the way the input is specified and how it differs from ordinary text. It is clear that AI techniques of syntactic and semantic analysis are necessary to achieve the goal of generating natural speech from text.