Portal de Periódicos da CAPES

On the definition of a prosodically balaced corpus: combining greedy algorithms with expert guided manipulation

2009; Technical University of Valencia; Issue: 43 Linguagem: Inglês

1135-5948

David Escudero-Mancebo, Lourdes Aguilar, Antonio Bonafonte Cávez, Juan María Garrido Almiñana,

Speech Recognition and Synthesis

This article reports the process of building a balanced text corpus taking into account prosodic features. We formalize the application of greedy algorithms for text selection and we discuss their limitations. We also defend an expert guideline for text manipulation that significantly improves the performance of the algorithms. The application of this methodology to a radio news corpus empirically supports the proposed strategy.