5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Forming Generic Models Of Speech For Uniform Database Access

Toomas Altosaar (1), Martti Vainio (2)

(1) Helsinki University of Technology, Finland
(2) University of Helsinki, Finland

This paper presents a formalism that models speech from different databases generically. For each utterance in a speech database a communication framework is first constructed which is composed of a set of communication planes, such as acoustic, orthographic, linguistic, and phonetic. Each plane in turn is made up of a set of levels to represent the plane's structural hierarchy, e.g., for the linguistic plane, levels such as sentence, word, syllable, and phoneme may exist. Information from speech databases is parsed and compiled into such objects and exhibit both individual and class inherited behaviour. Once placed into the framework these objects can have their relationships to other objects explicitly defined by links on the same level, across different levels, and across different planes. Speech from databases covering different languages and annotation styles can therefore be modelled generically allowing for uniform database access. Searches can be performed on the framework and the results used for further analyses.

Full Paper

Bibliographic reference.  Altosaar, Toomas / Vainio, Martti (1998): "Forming generic models of speech for uniform database access", In ICSLP-1998, paper 0887.