Word Recognition and Speaker Authentication Using Amplitude Independent and Time Independent Word Features.

摘要

In two related speech research projects, a digital computer was programmed to recognize words and to authenticate speakers by mean of a very simple analysis of speech data. Recognition is based upon binary patterns generated by analysis of amplitude independent and time independent spectral features. The system provides reliable word recognition for vocabulary sizes up to 30words,provided that the speaker furnishes his own training data. Since the system is highly speaker sensitive,it also provides an excellent method of speaker authentication based on the ability of an 'unknown'speaker to produce patterns similar to those of a 'reference'speaker. No authentication errors were made in an experiment involving 20speakers. (Author)