Speech Acts and Prosodic Modeling in Text to Speech Systems
Valbon Ademi, University of Tetova (Macedonia, The former Yugoslav Republic of)
Lindita Ademi, University of Tetova (Macedonia, The former Yugoslav Republic of)
Abstract
In our introduction to the nature of language, we explained that one of its fundamental properties is that there is an arbitrary relation between the meaning and form of a word. A word’s pronunciation cannot be inferred from its meaning, nor vice versa. Obviously for a language to work, these arbitrary relations need to be somehow stored, and the device used to do this is call a lexicon.We can talk about a speaker’s lexicon, meaning the lexicon that a particular individual uses to store his or her words; we can also talk about the language lexicon which is a store of all the meaning/form pairs (words) in that language (a traditional dictionary is this type of lexicon). For language engineering purposes however, we talk about the computer lexicon, which is a physical computer data object containing known descriptions of words. In machine understanding and generation systems, or any systems which deal with semantics and meaning, the lexicon is a vital component in a system as it is here that the connection between meaning and spoken or written form is stored. In speech synthesis and recognition tasks however, the meaning element is largely ignored, so in principle the lexicon is not strictly speaking a necessary component. However, lexicons are in fact widely used in synthesis and recognition to store the correspondences between the two main forms of the word; that is the spelling and pronunciation of a word.
Keywords: Lexicon, lexicology, lexicography, terminology, speech recognition, Text-to-Speech;