Process for ascertaining a series of words and arrangement for this process

摘要

Electric signals generated from the speech signal are sampled at repeated intervals and a series of test signals is generated. The test signals are compared signal by signal with different series of reference signals to generate rating values. Each series of reference signals belongs to a predetermined set of reference signals and represents a word in the vocabulary. The set of steams of reference signals forms a tree structure with a root. Each branch of the tree has a number of reference signals and is allocated to a speech element. Vocabulary words are allocated to determined branch nodes and branch ends. A word result is provided for each finished word. The result includes a total rating value. The comparison with generation of rating values is contained, starting with a new rating value corresponding to the total rating, for successive test signals with the start of a stream of reference signals, as long as the rating value is less than a given threshold. The total rating of the finished word is taken from the rating at the end of this word and from a speech model value. At least one stream of words is generated on the basis of the total rating. As the comparison is continued to the end of a word, the rating value is increased by a correction value, at least at the root of the tree structure. When the end of the word is reached, this correction value is subtracted from the rating value at the word end and is added to the speech model to provide a total rating value.