Query Language for Access to Speech Corpora

With more and more speech corpora at hand the unit selection technique is a promising approach in concatenative speech synthesis. What is missing are models of optimal parameters that sufficiently describe utterances to be produced and their corresponding counterparts in collections of speech data. Prior to this, existing corpora have to be annotated on possibly relevant linguistic and signal levels. This paper deals with standards developed in the MATE project for the uniform annotation of speech corpora to be represented in XML and a query language which can access these corpora. These standards may accelerate the identification of optimal elements for the annotation and description of parameters relevant for the unit selection technique.