In this paper we present an improved method for the spectral estimation of car noise in order to enhance the performance of noise reduction systems. An algorithm is developed that allows us to track changes in the noise spectrum during speech activity. For this tracking, we use the knowledge of the speed of the car and the revolutions of the engine. The paper starts with a detailed analysis of the car noise. The proposed algorithm based on this analysis first removes the harmonic components of the engine noise by selective filtering in time. The remaining wind and tyre noise is predicted during speech activity, based on the last available estimate and the vehicle speed.
展开▼