NaturTtSMLUn esquema de anotación para la mejora de la naturalidad en los sistemas de síntesis de voz
- Albert González Lamaña 1
- Antonio Pareja Lora 2
-
1
Universidad Nacional de Educación a Distancia
info
-
2
Universidad Complutense de Madrid
info
ISSN: 2444-197X
Year of publication: 2018
Issue: 4
Pages: 375-390
Type: Article
More publications in: E-Aesla
Abstract
Even though the output of current Text-to-Speech (TtS) systems is usually quite comprehensible, this output is quite often too monotonic. One of the main causes for this problem is that a TtS normally cannot actually understand plain text. Thus, the TtS system must be provided with the way to read it, in order to supply a more natural and expressive speech. So far, some annotation languages and schemes have been developed towards this end; however, they are partial and/or focus in different areas of expressive speech. This paper presents an annotation scheme (NaturTtSML) that merges and makes interoperate all of them altogether in just one scheme.