Pre-emphasis and speech recognition

It is known that noise can significantly decrease the performance of a speech recognition system. To solve this problem, many speech processing algorithms have been developed. Most of them assume that the noise level is constant, or is to be evaluated in the course of the algorithm. The paper addresses a particular kind of noise: the type introduced by pre-emphasis of the speech signal. A new enhancement speech procedure is presented which allows a more relevant spectrum without the need for an a priori knowledge of the noise intensity. It is shown that this approach can greatly increase the recognition rate.