| 摘要 |
<p>Signal (eg. sound or voice) sources (101, 102) are identified and separated by correlating artefacts detected at multiple sensors (eg. microphones 103-105) with source activity over time. Sources are first spatially localized (fig. 2) via eg. examining the density of peaks (figure 4) in a histogram representing accumulation over time of a 2D angular spectra curve specinst (θ,φ)(t) for horizontal angle θ and vertical angle φ, and then temporal correlations between localized source activity is determined by eg. calculating the Minimum Mean Square Error (MMSE) between an activity vs. time plot and its linear regression line, with an activity parameter Ei(t) being normalised and compared to thresholds (figure 6). Artefacts from indirect reception are less apparent in the time-averaged activity (figure 5).</p> |
