Apparatus and method for adapting model parameters of speech recognizer by utilizing histogram equalization

摘要

PURPOSE: An acoustic model parameter adaptation device using histogram equalization and a method thereof are provided to remove acoustic mismatch between test environment and drill environment, thereby preventing voice recognition performance which is strong against noise environment. CONSTITUTION: A test accumulative distribution function estimator(11) estimates test accumulative distribution function estimation about test voice feature parameters. A drill accumulative distribution function estimator(12) estimates drill accumulative distribution function estimation about a sound model average parameters. A linear interpolation factor calculator(13) obtains linear interpolation factor. A sound model parameter adapter(14) performs linear interpolation about the test voice feature parameter which the voice model average parameter performs through histogram equalization.