Apparatus and method for detection and classification of speech signals within broadband source signals

摘要

Within a broadband source signal for detecting and classifying speech signals, the source signal in time details, of each of the sections of the fourier transform as a column vector is determined, of each aperture vector d an auto correlation vector is formed, so that there is a two-dimensional spectral correlogram k contains. Subsequently, in the spectral correlogram k contains an amount of maximum positions, each characterized by a lines - and a columns indication, is determined and maximum positions are combined to maximum lists. Then, for each maximum list a description of the vector is formed, which consists of statistical values of the associated rows - and columns of information, maximum lists their description, vectors have values, the outside previously fixed limits are discarded, from the description of vectors of the not rejected maximum lists, statistical speech parameters are determined, and a positive classification message is then emitted if the statistical speech parameters within previously for this speech parameters predetermined threshold limits lie. In the case of a positive classification message is output a detection of message from the minimum and the maximum columns giving all is not rejected maximum lists.