7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

A Real-Time Acoustic Human-Machine Front-End for Multimedia Applications Integrating Robust Adaptive Beamforming and Stereophonic Acoustic Echo Cancellation

W. Herbordt (1), J. Ying (2), H. Buchner (1), W. Kellermann (1)

(1) University Erlangen-Nuremberg, Germany; (2) Intel Architecture Development Co. Ltd., China

Joint beamforming microphone arrays and multi-channel acoustic echo cancellation (AEC) can be efficiently applied for hands-free speech communication. Especially, systems relying on adaptive generalized sidelobe canceller (GSC) structures are very promising, since they combine high noise-reduction performance with computational efficiency. So far, robustness of the GSC was very challenging, due to reverberation and non-stationarity of desired signals and interferers. In this contribution, we present for the first time a real-time system which integrates GSC and stereophonic AEC. It is robust against desired signal cancellation while highly suppressing interference and acoustic echoes. The realization on a low-cost PC platform, with the microphone array connected directly to the universal serial bus (USB), provides maximum hardware and software compatibility for personalized mobile computing devices and desktop PCs.

Full Paper

Bibliographic reference.  Herbordt, W. / Ying, J. / Buchner, H. / Kellermann, W. (2002): "A real-time acoustic human-machine front-end for multimedia applications integrating robust adaptive beamforming and stereophonic acoustic echo cancellation", In ICSLP-2002, 773-776.