Safira - Affective Speech Module

Introduction

The Safira module for affective speech generation, will combine methods from two areas of language generation - the generation of written language and the synthesis of spoken language.

To accomplish the first task, we rely on a template-based generation approach that is parameterized by an agent's emotional state. To find an appropriate mapping between an agent's emotional state and the resulting natural-language utterances, we will adopt a corpus-based approach. For speech synthesis, we will use an existing Speech Synthesizer - probably Festival which will get the generated utterances enriched by intonational mark-ups as input.

Documents

  • ASM - Integration and Use (English, pdf, 365KB)
  • This document describes the functionality of the Affective Speech Module (ASM) and which interfaces it provides for integration into software and how it can be used in the SAFIRA Toolkit environment. Further on we present an example of use. This includes a rule set for generating utterances structures. Based on these utterances structures speech synthesisers are able to synthesise speech. The generation process depends on a representation on what to generate, the input structure, and on an affective state vector, which contains values for personality and emotions. This vector influences the format of utterance structures, like the content, syntactical word and phrase placement, and speech parameters.

Safira DFKI Main | Related Research | ASM Software | Related Software
Contact: Elisabeth André, Patrick Gebhard, © DFKI GmbH, Last modified: Mon Oct 7 10:53:15 CEST 2002