Clasificación de textos adaptada para Conversión de Texto en Habla Multidominio

This paper introduces a text classification system tuned to cope with the requirements of multi-domain text-to-speech synthesis. This method, based on a previous system which represents texts by means of a weighted graph, has been developed to improve the classification efficiency for small texts and to minimize its computational cost. To that effect, the comparison space is built from the input text instead of being built from the training documents. Classification experiments conducted on an advertising text corpus show the achievement of the posed goals.