Annotation and interpretation of prosodic data in the HuComTech corpus for multimodal user interfaces

Abstract

Prosody is an important component of multimodal communication. Appropriate interpretation of speech would be less effective and sometimes impossible without the information expressed by prosodic features. Complementing other sources of information, computer recognition of various communicative events and individual characteristics of the speaker are also based on prosody where measured physical features (such like fundamental frequency and intensity) can be used as parameters for further operations. This research paper is concerned with the differences of measured and perceived aspects of prosodic data and a prosodic annotation method which is implemented in the HuComTech corpus based on the results of existing applications and the psychoacoustic model of tonal perception.