Sound capture for human/machine interfaces. Practical aspects of microphone array signal processing (Q1768945)

Speech communication devices like speech recognizers work very well under laboratory conditions especially if the speech input is performed using a close-talk microphone. As soon as the distance between speaker and microphone is enlarged, however, the performance is degraded considerably due to some reasons. Among them, there are different types of noise sources which are frequently non-stationary, acoustic echoes from the output channels of the communication system, and the influences of the room which are summarized as reverberation and described by the room impulse response. It is the aim of a so-called acoustic front-end to minimize these influences which finally leads to a clear and undisturbed input signal. The book deals with the development of a specific front-end which includes an optimal combination of beamforming and acoustic echo cancellation with considerable mathematical depth. Following an introduction, chapter 2 introduces the mathematical description of space-time signals (acoustical wave fields) with deterministic and probabilistic methods. Chapter 3 applies the theory of optimum linear filters to linear MIMO (multiple-input multiple-output) systems. The chapter 4 is devoted to the discussion of space-time beamforming using microphone arrays. Performance measures for beamforming are introduced, and algorithms for data-independent and data-dependent beamformers are presented. Special attention is given to the GSC (Generalized Sidelobe Canceller) beamformer realization. This special solution is applied in an Audio Acquisition System in the following chapter. The robustness of this solution against different distortions is demonstrated by a number of experiments. In chapter 6, multi-channel strategies for cancelling acoustical echoes are combined with the beamforming algorithms, especially the GSC. The application of the solutions to embedded systems requires real-time algorithms which are consuming not too much computing power. This practical aspect is discussed in the last chapter. A nu mber of experiments confirm the suitability of the algorithms including the successful combination of the acoustic front-end with an HTK based speech recognizer. The very detailed and careful treatment recommends the book as a more general reference for acoustical signal processing in beamforming and echo cancellation applications.