Comparison of Speech Enhancement Systems for Noise Fields in a Car Environment

Authors:

Abstract:

For telecommunications in a car environment hand-sets have to be replaced by hands-free units due to an increasing demand for a convenient and natural dialog and for safety in road traffic. The clean speech signal is heavily corrupted by noise from the engine, the air conditioning etc. The noise field in a car is strongly low-pass filtered and can be instationary due to changing driving speeds or opening the window, e.g. Different multi-channel and single-channel speech enhancement algorithms will be compared for a car-environment in this contribution. Since the well-known Ephraim&Malah algorithm in connection with Martin's Minimum Statistics suppresses the noise without affecting the speech signal too much, multi-channel combinations of beamformers and post-filters can exploit spatial information and are therefore independent of the statistics of the noise field. In contrast many multi-channel post-filter algorithms suffer from poor noise reduction in the lower frequency ranges due to strongly correlated noise in the microphone channels. Hybrid multi-channel noise reduction schemes have the ability to combine the advantages of both systems. In correlated frequency ranges single channel Wiener-Filters or the Ephraim&Malah-algorithm can be used while the multi-channel post-filters exploit spatial information in the higher frequencies.

Document type:

Conference Paper

Publication:

Braunschweig, Germany, 20. - 23. March 2006

Conference:

32nd German Annual Conference on Acoustics (DAGA 06)

Pages:

45-46

Index:

262

Files:

Paper