UNSUPERVISED ADAPTATION OF A LARGE VOCABULARY AUTOMATIC SPEECH RECOGNIZER

摘要

A method of improving the recognition accuracy of a speech recognizer deployed in an environment to receive live input data. Live input data and associated recognition responses are collected and then used as inputs to a given speaker independent adaptation algorithm. The algorithm is then used to enhance the accuracy of the speech recognizer without human supervision. This technique is useful for adapting a large vocabulary ASR engine.