Process to ascertain a series of words.

摘要

Electric signals generated from the speech signal are sampled at repeated intervals and a series of test signals is generated. A first memory (14) accepts different series of reference signals, each representing a word of the vocabulary. The sampler (12) and memory (14) are coupled between a comparator (16) which compares the test signals with difference series of reference signals and generates rating values. A generator (18) coupled to the comparator (16) generates a word result when the end of a series of reference signals is reached, corresponding to the end of a word. Each word result includes a reference to the rating value reached, a reference to the associated word and a reference to the preceding word. A second memory (20) stores the word result. A third memory (22) stores words of a speech model. An arrangement (24) coupled to the second and third memories updates the rating value in each word result by a speech model value in the third memory (22), and selects the word result with the best rating value and supplies this to the comparator (16). At least one word stream is selected (26, 28) in predetermined time intervals, corresponding to a series of selected word results with good rating values and the selected streams are output.