Reusing Language Resources for Speech Applications involving Emotion
Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC 2004)
Abstract
The present paper involves using a spoken corpus for the construction of a written corpus which in turn will be used for speech applications involving emotion, namely an emotional Text-to-Speech system or a Speech-to-Emotion system which requires emotional speech recognition and consequent text to emotion conversion. Such speech application systems, involve the construction of a corpus of written artificial dialogs, which is intended to depict/convey the speakers emotional state. From the analysis of the sublanguage of the spoken corpus as well as the analysis and evaluation of the sublanguage of the (constructed) written corpus, we classified the extra-linguistic markers into three types ('Pause', 'Emphasis' and 'Hesitation') and we observe that, in Greek, extra-linguistic markers may behave as pointers to key-information. The correspondence between extra-linguistic markers and key-information was evaluated with an additional spoken corpus of recorded dialogs from selected Greek TV programs. The written corpus containing the inserted extra-linguistic markers was evaluated by native speakers with linguistic knowledge. The speakers were asked to classify them according to their acceptability in Greek (Alexandris & Fotinea, 2003).