Databases of Expressive Speech

This paper discusses the construction of speech databases for research into speech information processing and describes a problem illustrated by the case of emotional speech synthesis. It introduces a project for the processing of expressive speech, and describes the data collection techniques and the subsequent analysis of supra-linguistic, and emotional features signalled in the speech. It presents annotation guidelines for distinguishing speaking-style differences, and argues that the focus of analysis for expressive speech processing applications should be on the speaker relationships (defined herein), rather than on emotions.