Speech recognition system

摘要

In a speech recognition system, a vector quantisation method is used so that each frame of a time-frequency distribution produced by a filter bank from speech frequency analysis is converted into a vector-quantised code with reference to a previously created code book containing a set of phonemes of very specific speech. A distribution pattern for the frequency of each of the codes in the code book is created from these vector-quantised codes. A distribution pattern of this type is stored as reference data in a memory for a desired number of speech types when the system is operated in the registration mode. When the system is operated in a recognition mode, unknown speech is subjected to a frequency analysis and vector quantisation steps in order to create an unknown distribution pattern which is then compared with each of the reference data in order to identify the pattern to which this unknown speech corresponds.