Abstract: In communication networks with transmission devices using analog circuitry, because of unadapted 4-wire/2-wire junctions, some of the speech of a local speaker is reflected back to the local speaker as a line echo. This line echo is neutralized by echo compensators. However, updating of the filter coefficients of the echo compensators cannot be allowed to occur unless speech of the far-end speaker is absent. Thus, recalculation of the filter coefficients must be precluded at other times. Various possibilities for detecting the speech of a local speaker are well known. According to the invention, detection of the speech of a far-end speaker is fed back to obtain an attenuation measurement of the transmission path; this received speech simulates a reduction in attenuation. From the difference between the long-time average (Aa(k+1)) and the short-time average (A(k+1)) of the attenuation, it is learned very quickly, by comparison with a threshold (THRe), whether speech is present or not.

Abstract: During speech recognition of words, a precise and strong detection of start/end points of the words must be ensured, even in very noisy surroundings. Use of a feature with noise-resistant properties is shown wherein for a feature vector, a function of the signal energy is formed as the first feature and a function of the quadratic difference of an LPC (Linear-Predictive-Coding) cepstrum coefficient as a second feature. A check quantity or a maximum function of a distribution function is calculated, which detects the start/end points by comparison with a threshold.

Abstract: To determine the degree of correspondence between a first and a second pattern, the first and the second pattern are mapped to n first (V11 to V1n) and m second (V21 to V2m) feature vectors respectively. For points ({1, 1} to {n, m}) in a subarea of a matrix formed by the n first (V11 to V1n) and m second (V21 to V2m) feature vectors, the distance of the respective first (V11 to V1n) and the second (V21 to V2m) feature vectors is computed, and from the mean distance along an optimum path by means of a DP algorithm. Data regarding the gradient of the respective optimum path are determined during the computation for boundary points of the subarea, and the subarea is dynamically reduced for further computations on the basis of these data. The mean distance is used as the degree of correspondence.