Speech Enhancement for Hands-Free Car Phones by Adaptive Compensation of Harmonic Engine Noise Components

摘要

This paper presents a method for enhancing speech disturbed by car noise. The proposed method cancels the powerful harmonic components of engine noise by adaptive filtering which utilizes the known rpm signal available on the CAN bus in modern cars. The procedure can be used as a preprocessing method for classical broad-band noise reduction as it is able to cancel the engine noise - and thus a large amount of low-frequent car noise - without provoking speech distortion. The main part of the paper is dedicated to the step-size control of the utilized LMS algorithm necessary for a complete cancellation of the harmonics without speech distortion. Therefore, first a theoretically optimal step-size is determined and then a procedure is described which allows its determination in real applications. The paper concludes with a presentation of results obtained with this approach.