Summary and Conclusions
Safety and convenience issues require hands-free speech-based human— machine interfaces to manipulate complex functionalities and devices, for example, in cars. Such interfaces severely suffer from local interferences, such as the codriver voice, which have to be suppressed. As the desired signal and the local interference have the same nature (speech), it is difficult to separate them from their temporal (or spectral) properties. However, since they are emitted from different locations, separation of the desired signal may be resolved by exploiting the spatial properties of the source signals using microphone arrays.
A particularity of the car environment is that the position of the speakers relative to the array is known a priori. This prior information may be used directly using a linearly constrained minimum variance (LCMV) beamformer. Adaptive LCMV beamformers are able to attain a high suppression of the interference signal. Unfortunately they may also cancel the desired signal if the adaptation occurs during target signal activity. This target signal cancelation problem is a central motivation for the further investigations developed in this book.
KeywordsBlind Source Separation Natural Gradient Microphone Signal Local Interference Blind Source Separation Algorithm
Unable to display preview. Download preview PDF.