Description:
Sound capturing in an automotive environment without using a close-talk microphone is a challenging problem. First of all, the noise level inside a car is significantly higher than the noise level in a typical office room, resulting in an extremely low signal-to-noise ratio. Secondly, there exist various noise sources (passing cars, engine noise, whirling wind, and so on) which make the ambient noise highly non-stationary. As a result, conventional noise suppressing schemes - that have been mainly developed for stationary noise - do not give satisfactory results, and often tend to introduce undesirable artifacts. Generally, using a microphone array instead of a single microphone can improve the quality of the captured signal, as it allows us to do spatial filtering in addition to the traditional temporal filtering. In this talk, we propose potential techniques based on microphone arrays, which can be used for building good sound capturing systems in automotives. We consider various modifications of the traditional GSC (generalized sidelobe canceller), some which incorporate the estimates of instantaneous direction of arrival (IDOA) for improving the quality of the captured signal. We present the experimental results of various schemes (including the result of the conventional GSC) at the end of the talk, which indicates that the proposed structure is indeed very promising.
Speaker(s):
Byung-Jun Yoon, Ph.D. candidate, Department of Electrical Engineering, California Institute of Technology (Caltech)
|
