Correlation form of timestamp increment sequences of self-similar traffic on ethernet - Electronics Letters

rithm, we show a numerical experiment. A uniform linear array of 10 sensors is used. The sensor spacing is half of the wavelength corresponding to the upper band edge of the speech. The sampling frequency is 16kHz, and the frame length is 512 samples. The target speaker moves in constant angular velocity from -20 to 40”, while the interfering speaker is stationary at 0”. Fig. 2 shows the original speech waveform, the waveform produced by a futed beamformer (pointing at O”), and the waveform produced by the adaptive tracking beamformer. Significant improvement in audibility is demonstrated by Fig. 2. Subjectively, the output of the latter is much clearer than that of the fmed beam. Animations which show the beam pattern at each update (each frame) also demonstrate very good tracking capability. Of course, when the DOAs of the target and interfering speaker coincide, or are very close to each other, the DOAs cannot be identified and the signals cannot be separated. However, in general, the proposed method for tracking a spatially non-stationary speech source in the presence of an interfering speech source and low stationary background noise is fairly effective. The major problem is the computational complexity associated with the cyclostationarity feature extraction process, and work on the cyclostationarity of speech is required to provide a solid theoretical foundation.