Relational vs. object-oriented models for representing speech: a comparison using ANDOSL data

Much effort has been expended by the speech community in designing and producing speech corpora and databases. However, considerably less thought has been given to the type of database model that best represents speech for optimal database access. This paper reviews an established database model, the relational database management system (RDBMS), as well as an emerging one, the object-oriented database management system (OODBMS). Both model types are compared by addressing identical queries to the same data that is separately represented under these respective systems. This is accomplished using an identical source of richly described speech data from ANDOSL (Australian National Database of Spoken Language). By identifying the strengths and weaknesses of both approaches new models can be defined that incorporate the best of both existing systems enabling more fluent and extensive extraction of knowledge from speech databases.