An algorithm for locating the beginning and end of an utterance using adpcm coded speech

We describe a simple algorithm for locating the beginning and end of a speech utterance. The algorithm is based on the fact that the code words for an adaptive differential (ADPCM) representation of speech exhibit considerable variation among all quantization levels during both voiced and unvoiced speech intervals while, because of a constraint on the minimum step size, during silent intervals the code words vary only slightly within the smallest quantization steps. The use of the algorithm is illustrated for automatically locating the beginning and end of vocabulary entries for a computer voice response system.