An Efficient Representation of k-TSS Language Models

INÉS TORRES, AMPARO VARONA

Resumen


K-TESTABLE LANGUAGES IN THE STRICT SENSE (K-TSS), WHICH ARE A SUBCLASS OF REGULAR LANGUAGES, ARE USED IN THIS WORK TO BE INTEGRATED IN A CONTINUOUS SPEECH RECOGNITION (CSR) SYSTEM. AN EFFICIENT REPRESENTATION OF THE CORRESPONDING STOCHASTIC FINITE STATE AUTOMATON (SFSA), THAT INTEGRATES K K- TSS MODELS INTO A SELF-CONTAINED MODEL, IS PROPOSED IN THIS WORK. THE WHOLE MODEL IS REPRESENTED IN A SIMPLE ARRAY OF AN ADEQUATE SIZE THAT CAN BE EASILY HANDLED AT DECODING TIME BY A SIMPLE SEARCH FUNCTION. AN EXPERIMENTAL EVALUATION OF THE PROPOSED SFSA REPRESENTATION WAS CARRIED OUT OVER AN SPANISH RECOGNITION TASK. THESE EXPERIMENTS SHOWED THAT SYNTACTIC MODELS COULD BE EFFICIENTLY REPRESENTED AND INTEGRATED IN CSR SYSTEMS.

Palabras clave


CONTINUOUS SPEECH RECOGNITION; LANGUAGE MODELING; SYNTACTIC PATTERN RECOGNITION

Texto completo:

pdf


Contacto:
Oscar Zavala