PROCEDURE FOR THE VOICE RECOGNITION OF AN ANNOUNCER WHO PUTS IN PRACTICE A PREDICTIVE MODEL ESPECIALLY FOR ACCESS CONTROL APPLICATIONS.

摘要

THE INVENTION REFERS TO A VOICE RECOGNITION PROCEDURE OF A LOCUTOR WHO USES A PREDICTIVE MODEL OF ORDER Q, WHICH INCLUDES A PHASE OF EXTRACTION OF STATISTICAL CHARACTERISTICS THAT BRINGS A STAGE DIGITAL ACQUISION TO A SAMPLE OR SEVERAL ELOCUTIONS, A STAGE OF TRANSFORMATION OF THIS SOUND SAMPLE INTO A SERIES OF VECTORS OF DIMENSION P, OBTAINED FROM A SERIES OF WINDOWS OF MEDIUM DIMENSION TY AND OF MIDDLE INTERVAL I, AND A STAGE OF DETERMINATION OF Q + 1 MATRICES OF CORRELATION FROM THIS SERIES OF VECTORS. ACCORDING TO THE INVENTION, THE AVERAGE DIMENSION T HAS A DURATION LESS THAN 10 MS AND THE AVERAGE INTERVAL I A LENGTH LESS THAN 4.5 MS. PARTICULARLY APPLICATION IN A VOICE LOCK COMPRISING A SYSTEM OF ELECTROACOUSTIC CONVERSION (HP, 6) AND A SYSTEM (5) OF CARRYING OUT THE PROCEDURE WITH ENGRAVED PROGRAM.