Redefining Concatenative Speech Synthesis for Use in Spontaneous Conversational Dialogues; A Study with the GBO Corpus

This chapter describes how a very large corpus of conversational speech is being tested as a source of units for concatenative speech synthesis. It shows that the challenge no longer lies in phone-sized unit selection, but in categorising larger units for their affective and pragmatic effect. The work is by nature exploratory, but much progress has been achieved and we now have the beginnings of an understanding of the types of grammar and the ontology of vocal productions that will be required for the interactive synthesis of conversational speech. The chapter describes the processes involved and explains some of the features selected for optimal expressive speech rendering